GitHub/mt8127/android_kernel_alcatel_ttab.git
11 years agomac802154: Increase tx_buffer_len
Alan Ott [Wed, 3 Apr 2013 04:00:57 +0000 (04:00 +0000)]
mac802154: Increase tx_buffer_len

Increase the buffer length from 10 to 300 packets. Consider that traffic on
mac802154 devices will often be 6LoWPAN, and a full-length (1280 octet)
IPv6 packet will fragment into 15 6LoWPAN fragments (because the MTU of
IEEE 802.15.4 is 127).  A 300-packet queue is really 20 full-length IPv6
packets.

With a queue length of 10, an entire IPv6 packet was unable to get queued
at one time, causing fragments to be dropped, and making reassembly
impossible.

Signed-off-by: Alan Ott <alan@signal11.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agomac802154: Use netif flow control
Alan Ott [Wed, 3 Apr 2013 04:00:56 +0000 (04:00 +0000)]
mac802154: Use netif flow control

Use netif_stop_queue() and netif_wake_queue() to control the flow of
packets to mac802154 devices.  Since many IEEE 802.15.4 devices have no
output buffer, and since the mac802154 xmit() function is designed to
block, netif_stop_queue() is called after each packet.

Signed-off-by: Alan Ott <alan@signal11.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agomac802154: Do not try to resend failed packets
Alan Ott [Wed, 3 Apr 2013 04:00:55 +0000 (04:00 +0000)]
mac802154: Do not try to resend failed packets

When ops->xmit() fails, drop the packet. Devices which support hardware
ack and retry (which include all devices currently supported by mainline),
will automatically retry sending the packet (in the hardware) up to 3
times, per the 802.15.4 spec.  There is no need, and it is incorrect to
try to do it in mac802154.

Signed-off-by: Alan Ott <alan@signal11.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agosctp: fix error return code in __sctp_connect()
Wei Yongjun [Wed, 3 Apr 2013 03:02:28 +0000 (03:02 +0000)]
sctp: fix error return code in __sctp_connect()

Fix to return a negative error code from the error handling
case instead of 0, as returned elsewhere in this function.

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Acked-by: Vlad Yasevich <vyasevich@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years ago802: fix a possible race condition
Cong Wang [Tue, 2 Apr 2013 21:52:40 +0000 (21:52 +0000)]
802: fix a possible race condition

(Resend with a better changelog)

garp_pdu_queue() should ways be called with this spin lock.
garp_uninit_applicant() only holds rtnl lock which is not
enough here.  A possible race can happen as garp_pdu_rcv()
is called in BH context:

garp_pdu_rcv()
  |->garp_pdu_parse_msg()
    |->garp_pdu_parse_attr()
      |-> garp_gid_event()

Found by code inspection.

Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: David Ward <david.ward@ll.mit.edu>
Cc: "Jorge Boncompte [DTI2]" <jorge@dti2.net>
Signed-off-by: Cong Wang <amwang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoselftests: net: add PF_PACKET TPACKET v1/v2/v3 selftests
Daniel Borkmann [Tue, 2 Apr 2013 13:00:51 +0000 (13:00 +0000)]
selftests: net: add PF_PACKET TPACKET v1/v2/v3 selftests

This patch adds a simple test case that probes the packet socket's
TPACKET_V1, TPACKET_V2 and TPACKET_V3 behavior regarding mmap(2)'ed
I/O for a small burst of 100 packets. The test currently runs for ...

  TPACKET_V1: RX_RING, TX_RING
  TPACKET_V2: RX_RING, TX_RING
  TPACKET_V3: RX_RING

... and will output on success:

  test: TPACKET_V1 with PACKET_RX_RING .................... 100 pkts (9600 bytes)
  test: TPACKET_V1 with PACKET_TX_RING .................... 100 pkts (9600 bytes)
  test: TPACKET_V2 with PACKET_RX_RING .................... 100 pkts (9600 bytes)
  test: TPACKET_V2 with PACKET_TX_RING .................... 100 pkts (9600 bytes)
  test: TPACKET_V3 with PACKET_RX_RING .................... 100 pkts (9600 bytes)
  OK. All tests passed

Reusable parts of psock_fanout.c have been put into a psock_lib.h
file for common usage. Test case successfully tested on x86_64.

Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agovxlan: Bypass encapsulation if the destination is local
Sridhar Samudrala [Tue, 2 Apr 2013 12:31:52 +0000 (12:31 +0000)]
vxlan: Bypass encapsulation if the destination is local

This patch bypasses vxlan encapsulation if the destination vxlan
endpoint is a local device.

Changes since v1: added missing check for vxlan_find_vni() failure

Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet/mlx4_en: Advertize DCB_CAP_DCBX_HOST in getdcbx
Or Gerlitz [Sun, 7 Apr 2013 03:44:08 +0000 (03:44 +0000)]
net/mlx4_en: Advertize DCB_CAP_DCBX_HOST in getdcbx

When our getdcbx entry is called, DCB_CAP_DCBX_HOST should be advertized too.

Cc: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet/mlx4_en: Enable DCB ETS ops only when supported by the firmware
Or Gerlitz [Sun, 7 Apr 2013 03:44:07 +0000 (03:44 +0000)]
net/mlx4_en: Enable DCB ETS ops only when supported by the firmware

Enable the DCB ETS ops only when supported by the firmware. For older firmware/cards
which don't support ETS, advertize only PFC DCB ops.

Signed-off-by: Eugenia Emantayev <eugenia@mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet/mlx4_core: Added proper description for two device capabilities
Or Gerlitz [Sun, 7 Apr 2013 03:44:06 +0000 (03:44 +0000)]
net/mlx4_core: Added proper description for two device capabilities

Added readable description for the DPDP and port sensing device capabilities.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: phy: spi_ks8995: use spi_get_drvdata() and spi_set_drvdata()
Jingoo Han [Fri, 5 Apr 2013 20:37:07 +0000 (20:37 +0000)]
net: phy: spi_ks8995: use spi_get_drvdata() and spi_set_drvdata()

Use the wrapper functions for getting and setting the driver data using
spi_device instead of using dev_{get|set}_drvdata with &spi->dev, so we
can directly pass a struct spi_device.

Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: wireless: p54spi: use spi_get_drvdata() and spi_set_drvdata()
Jingoo Han [Fri, 5 Apr 2013 20:36:47 +0000 (20:36 +0000)]
net: wireless: p54spi: use spi_get_drvdata() and spi_set_drvdata()

Use the wrapper functions for getting and setting the driver data using
spi_device instead of using dev_{get|set}_drvdata with &spi->dev, so we
can directly pass a struct spi_device.

Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: wireless: wl1251: use spi_get_drvdata() and spi_set_drvdata()
Jingoo Han [Fri, 5 Apr 2013 20:36:25 +0000 (20:36 +0000)]
net: wireless: wl1251: use spi_get_drvdata() and spi_set_drvdata()

Use the wrapper functions for getting and setting the driver data using
spi_device instead of using dev_{get|set}_drvdata with &spi->dev, so we
can directly pass a struct spi_device.

Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: ethernet: enc28j60: use spi_get_drvdata() and spi_set_drvdata()
Jingoo Han [Fri, 5 Apr 2013 20:36:11 +0000 (20:36 +0000)]
net: ethernet: enc28j60: use spi_get_drvdata() and spi_set_drvdata()

Use the wrapper functions for getting and setting the driver data using
spi_device instead of using dev_{get|set}_drvdata with &spi->dev, so we
can directly pass a struct spi_device.

Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: ethernet: ks8851: use spi_get_drvdata() and spi_set_drvdata()
Jingoo Han [Fri, 5 Apr 2013 20:35:43 +0000 (20:35 +0000)]
net: ethernet: ks8851: use spi_get_drvdata() and spi_set_drvdata()

Use the wrapper functions for getting and setting the driver data using
spi_device instead of using dev_{get|set}_drvdata with &spi->dev, so we
can directly pass a struct spi_device.

Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: can: mcp251x: use spi_get_drvdata() and spi_set_drvdata()
Jingoo Han [Fri, 5 Apr 2013 20:35:14 +0000 (20:35 +0000)]
net: can: mcp251x: use spi_get_drvdata() and spi_set_drvdata()

Use the wrapper functions for getting and setting the driver data using
spi_device instead of using dev_{get|set}_drvdata with &spi->dev, so we
can directly pass a struct spi_device.

Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: ieee802154: mrf24j40: use spi_get_drvdata() and spi_set_drvdata()
Jingoo Han [Fri, 5 Apr 2013 20:34:18 +0000 (20:34 +0000)]
net: ieee802154: mrf24j40: use spi_get_drvdata() and spi_set_drvdata()

Use the wrapper functions for getting and setting the driver data using
spi_device instead of using dev_{get|set}_drvdata with &spi->dev, so we
can directly pass a struct spi_device.

Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agor8169: add a new chip for RTL8106E
hayeswang [Mon, 1 Apr 2013 22:23:42 +0000 (22:23 +0000)]
r8169: add a new chip for RTL8106E

Add a new chip for RTL8106E series.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agor8169: adjust the flow of hw_start
hayeswang [Mon, 1 Apr 2013 22:23:41 +0000 (22:23 +0000)]
r8169: adjust the flow of hw_start

The suggestion as following:
- initial settings or default settings
- rtl_hw_start_xxx. rtl_hw_start_xxx may change some default settings.
- enable tx/rx. This has to be after the above two steps.
- rtl_set_rx_mode. AcceptXXXs have to be enabled after enabling tx/rx.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agor8169: add a new chip for RTL8111G
hayeswang [Mon, 1 Apr 2013 22:23:40 +0000 (22:23 +0000)]
r8169: add a new chip for RTL8111G

Add a new chip for RTL8111G series.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agor8169: Update the RTL8111G parameters
hayeswang [Mon, 1 Apr 2013 22:23:39 +0000 (22:23 +0000)]
r8169: Update the RTL8111G parameters

- replace rtl8168g-1.fw with rtl8168g-2.fw which support new method.
- fix PHY power down is useless.
- disable rx early which causes the rx abnormal.
- enable auto fifo.
- set 10M IFG to default value.
- fix the conflict between jumbo frame and flow control.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agor8169: Modify the method for setting firmware
hayeswang [Mon, 1 Apr 2013 22:23:38 +0000 (22:23 +0000)]
r8169: Modify the method for setting firmware

Remove useless action PHY_READ_EFUSE, PHY_READ_MAC_BYTE, PHY_WRITE_MAC_BYTE,
PHY_WRITE_ERI_WORD. And define the new action PHY_MDIO_CHG.

PHY_MDIO_CHG is used to modify the mdio operation. By the way, the
firmware could support setting mac ocp.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agor8169: Update PHY settings of RTL8111G
hayeswang [Mon, 1 Apr 2013 22:23:37 +0000 (22:23 +0000)]
r8169: Update PHY settings of RTL8111G

Add the new settings and correct the wrong settings.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agor8169: Modify the mothod for PHY settings of RTL8111G
hayeswang [Mon, 1 Apr 2013 22:23:36 +0000 (22:23 +0000)]
r8169: Modify the mothod for PHY settings of RTL8111G

Replace the current settings with rtl_writephy and rtl_readphy.
For the hardware, the settings are same with previous ones. This
make the setting method like the previous chips.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agor8169: Remove firmware code
hayeswang [Mon, 1 Apr 2013 22:23:35 +0000 (22:23 +0000)]
r8169: Remove firmware code

Some codes are belong to binary codes and should be removed.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'master' of git://1984.lsi.us.es/nf-next
David S. Miller [Sun, 7 Apr 2013 16:22:06 +0000 (12:22 -0400)]
Merge branch 'master' of git://1984.lsi.us.es/nf-next

Pablo Neira Ayuso says:

====================
The following patchset contains Netfilter and IPVS updates for
your net-next tree, most relevantly they are:

* Add net namespace support to NFLOG, ULOG and ebt_ulog and NFQUEUE.
  The LOG and ebt_log target has been also adapted, but they still
  depend on the syslog netnamespace that seems to be missing, from
  Gao Feng.

* Don't lose indications of congestion in IPv6 fragmentation handling,
  from Hannes Frederic Sowa.i

* IPVS conversion to use RCU, including some code consolidation patches
  and optimizations, also some from Julian Anastasov.

* cpu fanout support for NFQUEUE, from Holger Eitzenberger.

* Better error reporting to userspace when dropping packets from
  all our _*_[xfrm|route]_me_harder functions, from Patrick McHardy.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonetfilter: implement RFC3168 5.3 (ecn protection) for ipv6 fragmentation handling
Hannes Frederic Sowa [Sun, 24 Mar 2013 15:20:51 +0000 (15:20 +0000)]
netfilter: implement RFC3168 5.3 (ecn protection) for ipv6 fragmentation handling

This change brings netfilter reassembly logic on par with
reassembly.c. The corresponding change in net-next is
(eec2e61 ipv6: implement RFC3168 5.3 (ecn protection) for
ipv6 fragmentation handling)

Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Jesper Dangaard Brouer <jbrouer@redhat.com>
Cc: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonetfilter: remove unneeded variable proc_net_netfilter
Pablo Neira Ayuso [Fri, 5 Apr 2013 17:40:10 +0000 (19:40 +0200)]
netfilter: remove unneeded variable proc_net_netfilter

Now that this supports net namespace for nflog and nfqueue,
we can remove the global proc_net_netfilter which has no
clients anymore.

Based on patch from Gao feng.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonetfilter: nfnetlink_queue: add net namespace support for nfnetlink_queue
Gao feng [Sun, 24 Mar 2013 23:50:47 +0000 (23:50 +0000)]
netfilter: nfnetlink_queue: add net namespace support for nfnetlink_queue

This patch makes /proc/net/netfilter/nfnetlink_queue pernet.
Moreover, there's a pernet instance table and lock.

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonetfilter: enable per netns support for nf_loggers
Gao feng [Fri, 5 Apr 2013 15:04:35 +0000 (17:04 +0200)]
netfilter: enable per netns support for nf_loggers

After this patch, all nf_loggers support net namespace. Still
xt_LOG and ebt_log require syslog netns support.

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonetfilter: nfnetlink_log: add net namespace support for nfnetlink_log
Gao feng [Sun, 24 Mar 2013 23:50:45 +0000 (23:50 +0000)]
netfilter: nfnetlink_log: add net namespace support for nfnetlink_log

This patch makes /proc/net/netfilter/nfnetlink_log pernet.
Moreover, there's a pernet instance table and lock.

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonetfilter: ipt_ULOG: add net namespace support for ipt_ULOG
Gao feng [Sun, 24 Mar 2013 23:50:44 +0000 (23:50 +0000)]
netfilter: ipt_ULOG: add net namespace support for ipt_ULOG

Add pernet support to ipt_ULOG by means of the new nf_log_set
function added in (30e0c6a netfilter: nf_log: prepare net
namespace support for loggers).

This patch also make ulog_buffers and netlink socket
nflognl per netns.

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonetfilter: ebt_ulog: add net namespace support for ebt_ulog
Gao feng [Sun, 24 Mar 2013 23:50:43 +0000 (23:50 +0000)]
netfilter: ebt_ulog: add net namespace support for ebt_ulog

Add pernet support to ebt_ulog by means of the new nf_log_set
function added in (30e0c6a netfilter: nf_log: prepare net
namespace support for loggers).

This patch also make ulog_buffers and netlink socket
ebtulognl per netns.

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonetfilter: xt_LOG: add net namespace support for xt_LOG
Gao feng [Sun, 24 Mar 2013 23:50:42 +0000 (23:50 +0000)]
netfilter: xt_LOG: add net namespace support for xt_LOG

Add pernet support to xt_LOG by means of the new nf_log_set
function added in (30e0c6a netfilter: nf_log: prepare net
namespace support for loggers).

Since syslog ns has yet not been implemented, we don't want
the containers to DDOS host's syslogd. So only enable ebt_log
only from init_net and wait for syslog ns support

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonetfilter: ebt_log: add net namespace support for ebt_log
Gao feng [Sun, 24 Mar 2013 23:50:41 +0000 (23:50 +0000)]
netfilter: ebt_log: add net namespace support for ebt_log

Add pernet support to ebt_log by means of the new nf_log_set
function added in (30e0c6a netfilter: nf_log: prepare net
namespace support for loggers).

Since syslog ns has yet not been implemented, we don't want
the containers to DDOS host's syslogd. So only enable ebt_log
only from init_net and wait for syslog ns support.

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonetfilter: nf_log: prepare net namespace support for loggers
Gao feng [Sun, 24 Mar 2013 23:50:40 +0000 (23:50 +0000)]
netfilter: nf_log: prepare net namespace support for loggers

This patch adds netns support to nf_log and it prepares netns
support for existing loggers. It is composed of four major
changes.

1) nf_log_register has been split to two functions: nf_log_register
   and nf_log_set. The new nf_log_register is used to globally
   register the nf_logger and nf_log_set is used for enabling
   pernet support from nf_loggers.

   Per netns is not yet complete after this patch, it comes in
   separate follow up patches.

2) Add net as a parameter of nf_log_bind_pf. Per netns is not
   yet complete after this patch, it only allows to bind the
   nf_logger to the protocol family from init_net and it skips
   other cases.

3) Adapt all nf_log_packet callers to pass netns as parameter.
   After this patch, this function only works for init_net.

4) Make the sysctl net/netfilter/nf_log pernet.

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonetfilter: make /proc/net/netfilter pernet
Gao feng [Sun, 24 Mar 2013 23:50:39 +0000 (23:50 +0000)]
netfilter: make /proc/net/netfilter pernet

This patch makes this proc dentry pernet. So far only init_net
had a /proc/net/netfilter directory.

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonet: frag queue per hash bucket locking
Jesper Dangaard Brouer [Wed, 3 Apr 2013 23:38:16 +0000 (23:38 +0000)]
net: frag queue per hash bucket locking

This patch implements per hash bucket locking for the frag queue
hash.  This removes two write locks, and the only remaining write
lock is for protecting hash rebuild.  This essentially reduce the
readers-writer lock to a rebuild lock.

This patch is part of "net: frag performance followup"
 http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:

Same test setup as previous:
 (http://thread.gmane.org/gmane.linux.network/257155)
 Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
 Ethernet flow-control.  A third interface is used for generating the
 DoS attack (with trafgen).

Notice, I have changed the frag DoS generator script to be more
efficient/deadly.  Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.

Test types summary (netperf UDP_STREAM):
 Test-20G64K     == 2x10G with 65K fragments
 Test-20G3F      == 2x10G with 3x fragments (3*1472 bytes)
 Test-20G64K+DoS == Same as 20G64K with frag DoS
 Test-20G3F+DoS  == Same as 20G3F  with frag DoS
 Test-20G64K+MQ  == Same as 20G64K with Multi-Queue frag DoS
 Test-20G3F+MQ   == Same as 20G3F  with Multi-Queue frag DoS

When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression.  BUT this
was caused by some unrelated change in-between.  See tests below.

Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch

Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.

Performance table summary (in Mbit/s):

(#) Test-type:  20G64K    20G3F    20G64K+DoS  20G3F+DoS  20G64K+MQ 20G3F+MQ
    ----------  -------   -------  ----------  ---------  --------  -------
(A) Patch-02  : 18848.7   13230.1   4103.04     5310.36     130.0    440.2
(B) 1b5ab0de  : 18841.5   13156.8   4101.08     5314.57     129.0    424.2
(C) Patch-03v1: 18838.0   13490.5   4405.11     6814.72     196.6    461.6

(D) a210576c  : 18321.5   11250.4   3635.34     5160.13     119.1    405.2
(E) with _bh  : 17247.3   11492.6   3994.74     6405.29     166.7    413.6
(F) without bh: 17471.3   11298.7   3818.05     6102.11     165.7    406.3

Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.

I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried).  But the other results does show
improvements.  And test (E) "with _bh" version is slightly better.

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
  need the spinlock _bh versions, as Netfilter currently does a
  local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Wed, 3 Apr 2013 05:31:54 +0000 (01:31 -0400)]
Merge git://git./linux/kernel/git/davem/net

Pull net into net-next to get the synchronize_net() bug fix in
bonding.

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Wed, 3 Apr 2013 01:58:01 +0000 (18:58 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Fix VSOCK layer handling of context ID changes, from Reilly Grant.

 2) Now that we have a synchronize_net() in netdev_rx_handler_unregister(),
    we can't let any call sites hold locks.  Unfortunately bonding does,
    so we have to drop the rwlock there a little bit earlier, fix from
    Veaceslav Falico.

 3) MAC address setting loop exits one iteration too early in mlx4
    driver, from Yan Burman.

 4) Restore ipv6 routes properly upon ifdown/ifup of loopback, from
    Balakumaran Kannan.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net:
  VSOCK: Handle changes to the VMCI context ID.
  net IPv6 : Fix broken IPv6 routing table after loopback down-up
  cbq: incorrect processing of high limits
  net/mlx4_en: Fix setting initial MAC address
  bonding: get netdev_rx_handler_unregister out of locks

11 years agoMerge tag 'regmap-v3.9-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie...
Linus Torvalds [Wed, 3 Apr 2013 01:53:43 +0000 (18:53 -0700)]
Merge tag 'regmap-v3.9-rc4' of git://git./linux/kernel/git/broonie/regmap

Pull regmap fixes from Mark Brown:
 "A small collection of fixes.  The most important ones are those from
  Stephen and Lars-Peter both of which fix cache issues that have been
  lurking for a while but not manifesting noticably enough for anyone to
  report them."

* tag 'regmap-v3.9-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap:
  regmap: async: Add missing return
  regmap: don't corrupt work buffer in _regmap_raw_write()
  regmap: cache Fix regcache-rbtree sync
  regmap: Initialize `map->debugfs' before regcache

11 years agoMerge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Wed, 3 Apr 2013 01:52:24 +0000 (18:52 -0700)]
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux

Pull DRM fixes from Dave Airlie:
 "Two core fixes, both regressions, along with some intel and some
  nouveau fixes for regressions and oopses"

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux:
  drm: correctly restore mappings if drm_open fails
  drm/nouveau: fix NULL ptr dereference from nv50_disp_intr()
  drm/nouveau: fix handling empty channel list in ioctl's
  drm: don't unlock in the addfb error paths
  drm/i915: Fix build failure
  drm/i915: Be sure to turn hsync/vsync back on at crt enable (v2)
  drm/i915: duct-tape locking when eDP init fails

11 years agoMerge branch 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus
Linus Torvalds [Wed, 3 Apr 2013 01:47:23 +0000 (18:47 -0700)]
Merge branch 'upstream' of git://git.linux-mips.org/ralf/upstream-linus

Pull MIPS fixes from Ralf Baechle:
 "A collection of fixes pretty much across the MIPS code.  Even the
  change to include/linux/signal.h by David Howells' 2a1486981c13 ("Fix
  breakage in MIPS siginfo handling") should be considered MIPS-specific
  as it touches an ifdefed segment that is only relevant to MIPS and
  which unfortunately can't be made to go away entirely."

* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus:
  Fix breakage in MIPS siginfo handling
  Revert "MIPS: BCM63XX: Call board_register_device from device_initcall()"
  MIPS: BCM63XX: Make nvram checksum failure non fatal
  MIPS: Fix code generation for non-DSP capable CPUs
  MIPS: Fix inconsistent formatting inside /proc/cpuinfo
  MIPS: SEAD3: Enable LL/SC.
  MIPS: Get rid of CONFIG_CPU_HAS_LLSC again
  MIPS: Add dependencies for HAVE_ARCH_TRANSPARENT_HUGEPAGE
  MIPS: VR4133: Fix probe for LL/SC.
  MIPS: Fix logic errors in bitops.c
  MIPS: Use CONFIG_CPU_MIPSR2 in csum_partial.S
  MIPS: compat: Return same error ENOSYS as native for invalid operation.

11 years agonet: fix smatch warnings inside datagram_poll
Jacob Keller [Tue, 2 Apr 2013 20:55:40 +0000 (13:55 -0700)]
net: fix smatch warnings inside datagram_poll

Commit 7d4c04fc170087119727119074e72445f2bb192b ("net: add option to enable
error queue packets waking select") has an issue due to operator precedence
causing the bit-wise OR to bind to the sock_flags call instead of the result of
the terniary conditional. This fixes the *_poll functions to work properly. The
old code results in "mask |= POLLPRI" instead of what was intended, which is to
only include POLLPRI when the socket option is enabled.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agodrm: correctly restore mappings if drm_open fails
Ilija Hadzic [Tue, 2 Apr 2013 13:56:10 +0000 (09:56 -0400)]
drm: correctly restore mappings if drm_open fails

If first drm_open fails, the error-handling path will
incorrectly restore inode's mapping to NULL. This can
cause the crash later on. Fix by separately storing
away mapping pointers that drm_open can touch and
restore each from its own respective variable if the
call fails.

Fixes: https://bugzilla.novell.com/show_bug.cgi?id=807850
(thanks to Michal Hocko for investigating investigating and
finding the root cause of the bug)

Reference:
http://lists.freedesktop.org/archives/dri-devel/2013-March/036564.html

v2: Use one variable to store file and inode mapping
    since they are the same at the function entry.
    Fix spelling mistakes in commit message.

v3: Add reference to the original bug report.

Reported-by: Marco Munderloh <munderl@tnt.uni-hannover.de>
Tested-by: Marco Munderloh <munderl@tnt.uni-hannover.de>
Signed-off-by: Ilija Hadzic <ihadzic@research.bell-labs.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: stable@vger.kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agoMerge branch 'drm-nouveau-fixes-3.9' of git://anongit.freedesktop.org/git/nouveau...
Dave Airlie [Tue, 2 Apr 2013 20:44:02 +0000 (06:44 +1000)]
Merge branch 'drm-nouveau-fixes-3.9' of git://anongit.freedesktop.org/git/nouveau/linux-2.6 into drm-next

Oops fixers.
* 'drm-nouveau-fixes-3.9' of git://anongit.freedesktop.org/git/nouveau/linux-2.6:
  drm/nouveau: fix NULL ptr dereference from nv50_disp_intr()
  drm/nouveau: fix handling empty channel list in ioctl's

11 years agonet/nxp/lpc_eth: Drop ifdef CONFIG_OF_NET
Guenter Roeck [Tue, 2 Apr 2013 09:35:11 +0000 (09:35 +0000)]
net/nxp/lpc_eth: Drop ifdef CONFIG_OF_NET

Since of_get_mac_address() is now declared even if CONFIG_OF_NET
is not configured, the ifdef is no longer necessary and can be
removed.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet/freescale/fec: Simplify OF dependencies
Guenter Roeck [Tue, 2 Apr 2013 09:35:10 +0000 (09:35 +0000)]
net/freescale/fec: Simplify OF dependencies

Since of_get_mac_address() is now defined even if CONFIG_OF_NET
is not configured, the ifdef around the code calling it is no longer
necessary and can be removed.

Similar, since of_get_phy_mode() is now defined as dummy function
if OF_NET is not configured, it is no longer necessary to provide
an OF dependent function as front-end. Also, the function depends
on OF_NET, not on OF, so the conditional code was not correct anyway.
Drop the front-end function and call of_get_phy_mode() directly.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet/cadence/macb: Simplify OF dependencies
Guenter Roeck [Tue, 2 Apr 2013 09:35:09 +0000 (09:35 +0000)]
net/cadence/macb: Simplify OF dependencies

With of_get_mac_address() and of_get_phy_mode() now defined as dummy
functions if OF_NET is not configured, it is no longer necessary to
provide OF dependent functions as front-end. Also, the two functions
depend on OF_NET, not on OF, so the conditional code was not correct
anyway.

Drop the front-end functions and call of_get_mac_address() and
of_get_phy_mode() directly instead.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet/cadence/at91_ether: Simplify OF dependencies
Guenter Roeck [Tue, 2 Apr 2013 09:35:08 +0000 (09:35 +0000)]
net/cadence/at91_ether: Simplify OF dependencies

With of_get_mac_address() and of_get_phy_mode() now defined as dummy
functions if OF_NET is not configured, it is no longer necessary to
provide OF dependent functions as front-end. Also, the two functions
depend on OF_NET, not on OF, so the conditional code was not correct
anyway.

Drop the front-end functions and call of_get_mac_address() and
of_get_phy_mode() directly instead.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoof_net.h: Provide empty functions if OF_NET is not configured
Guenter Roeck [Tue, 2 Apr 2013 09:35:07 +0000 (09:35 +0000)]
of_net.h: Provide empty functions if OF_NET is not configured

of_get_mac_address() and of_get_phy_mode() are only provided if OF_NET
is configured. While most callers check for the define, not all do, and those
who do require #ifdef around the code. For those who don't, the missing check
can result in errors such as

arch/powerpc/sysdev/tsi108_dev.c:107:3: error: implicit declaration of
function 'of_get_mac_address' [-Werror=implicit-function-declaration]
arch/powerpc/sysdev/mv64x60_dev.c:253:2: error: implicit declaration of
function 'of_get_mac_address' [-Werror=implicit-function-declaration]

Provide empty functions if OF_NET is not configured.
This is safe because all callers do check the return values.

Cc: David Daney <david.daney@cavium.com>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Acked-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel...
Dave Airlie [Tue, 2 Apr 2013 20:41:15 +0000 (06:41 +1000)]
Merge branch 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel into drm-next

One locking regression fix, and a couple of other i915 ones.

* 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel:
  drm: don't unlock in the addfb error paths
  drm/i915: Fix build failure
  drm/i915: Be sure to turn hsync/vsync back on at crt enable (v2)
  drm/i915: duct-tape locking when eDP init fails

11 years agoVSOCK: Handle changes to the VMCI context ID.
Reilly Grant [Mon, 1 Apr 2013 18:41:52 +0000 (11:41 -0700)]
VSOCK: Handle changes to the VMCI context ID.

The VMCI context ID of a virtual machine may change at any time. There
is a VMCI event which signals this but datagrams may be processed before
this is handled. It is therefore necessary to be flexible about the
destination context ID of any datagrams received. (It can be assumed to
be correct because it is provided by the hypervisor.) The context ID on
existing sockets should be updated to reflect how the hypervisor is
currently referring to the system.

Signed-off-by: Reilly Grant <grantr@vmware.com>
Acked-by: Andy King <acking@vmware.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet IPv6 : Fix broken IPv6 routing table after loopback down-up
Balakumaran Kannan [Tue, 2 Apr 2013 10:45:05 +0000 (16:15 +0530)]
net IPv6 : Fix broken IPv6 routing table after loopback down-up

IPv6 Routing table becomes broken once we do ifdown, ifup of the loopback(lo)
interface. After down-up, routes of other interface's IPv6 addresses through
'lo' are lost.

IPv6 addresses assigned to all interfaces are routed through 'lo' for internal
communication. Once 'lo' is down, those routing entries are removed from routing
table. But those removed entries are not being re-created properly when 'lo' is
brought up. So IPv6 addresses of other interfaces becomes unreachable from the
same machine. Also this breaks communication with other machines because of
NDISC packet processing failure.

This patch fixes this issue by reading all interface's IPv6 addresses and adding
them to IPv6 routing table while bringing up 'lo'.

==Testing==
Before applying the patch:
$ route -A inet6
Kernel IPv6 routing table
Destination                    Next Hop                   Flag Met Ref Use If
2000::20/128                   ::                         U    256 0     0 eth0
fe80::/64                      ::                         U    256 0     0 eth0
::/0                           ::                         !n   -1  1     1 lo
::1/128                        ::                         Un   0   1     0 lo
2000::20/128                   ::                         Un   0   1     0 lo
fe80::xxxx:xxxx:xxxx:xxxx/128  ::                         Un   0   1     0 lo
ff00::/8                       ::                         U    256 0     0 eth0
::/0                           ::                         !n   -1  1     1 lo
$ sudo ifdown lo
$ sudo ifup lo
$ route -A inet6
Kernel IPv6 routing table
Destination                    Next Hop                   Flag Met Ref Use If
2000::20/128                   ::                         U    256 0     0 eth0
fe80::/64                      ::                         U    256 0     0 eth0
::/0                           ::                         !n   -1  1     1 lo
::1/128                        ::                         Un   0   1     0 lo
ff00::/8                       ::                         U    256 0     0 eth0
::/0                           ::                         !n   -1  1     1 lo
$

After applying the patch:
$ route -A inet6
Kernel IPv6 routing
table
Destination                    Next Hop                   Flag Met Ref Use If
2000::20/128                   ::                         U    256 0     0 eth0
fe80::/64                      ::                         U    256 0     0 eth0
::/0                           ::                         !n   -1  1     1 lo
::1/128                        ::                         Un   0   1     0 lo
2000::20/128                   ::                         Un   0   1     0 lo
fe80::xxxx:xxxx:xxxx:xxxx/128  ::                         Un   0   1     0 lo
ff00::/8                       ::                         U    256 0     0 eth0
::/0                           ::                         !n   -1  1     1 lo
$ sudo ifdown lo
$ sudo ifup lo
$ route -A inet6
Kernel IPv6 routing table
Destination                    Next Hop                   Flag Met Ref Use If
2000::20/128                   ::                         U    256 0     0 eth0
fe80::/64                      ::                         U    256 0     0 eth0
::/0                           ::                         !n   -1  1     1 lo
::1/128                        ::                         Un   0   1     0 lo
2000::20/128                   ::                         Un   0   1     0 lo
fe80::xxxx:xxxx:xxxx:xxxx/128  ::                         Un   0   1     0 lo
ff00::/8                       ::                         U    256 0     0 eth0
::/0                           ::                         !n   -1  1     1 lo
$

Signed-off-by: Balakumaran Kannan <Balakumaran.Kannan@ap.sony.com>
Signed-off-by: Maruthi Thotad <Maruthi.Thotad@ap.sony.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoipconfig: add informative timeout messages while waiting for carrier
Paul Gortmaker [Tue, 2 Apr 2013 13:58:25 +0000 (09:58 -0400)]
ipconfig: add informative timeout messages while waiting for carrier

Commit 3fb72f1e6e6165c5f495e8dc11c5bbd14c73385c ("ipconfig wait
for carrier") added a "wait for carrier on at least one interface"
policy, with a worst case maximum wait of two minutes.

However, if you encounter this, you won't get any feedback from
the console as to the nature of what is going on.  You just see
the booting process hang for two minutes and then continue.

Here we add a message so the user knows what is going on, and
hence can take action to rectify the situation (e.g. fix network
cable or whatever.)  After the 1st 10s pause, output now begins
that looks like this:

Waiting up to 110 more seconds for network.
Waiting up to 100 more seconds for network.
Waiting up to 90 more seconds for network.
Waiting up to 80 more seconds for network.
...

Since most systems will have no problem getting link/carrier in the
1st 10s, the only people who will see these messages are people with
genuine issues that need to be resolved.

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoforcedeth: Do a dma_mapping_error check after skb_frag_dma_map
Neil Horman [Mon, 1 Apr 2013 04:31:58 +0000 (04:31 +0000)]
forcedeth: Do a dma_mapping_error check after skb_frag_dma_map

This backtrace was recently reported on a 3.9 kernel:

Actual results: from syslog /var/log/messsages:
kernel: [17539.340285] ------------[ cut here ]------------
kernel: [17539.341012] WARNING: at lib/dma-debug.c:937 check_unmap+0x493/0x960()
kernel: [17539.341012] Hardware name: MS-7125
kernel: [17539.341012] forcedeth 0000:00:0a.0: DMA-API: device driver failed to
check map error[device address=0x0000000013c88000] [size=544 bytes] [mapped as
page]
kernel: [17539.341012] Modules linked in: fuse ebtable_nat ipt_MASQUERADE
nf_conntrack_netbios_ns nf_conntrack_broadcast ip6table_nat nf_nat_ipv6
ip6table_mangle ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 iptable_nat
nf_nat_ipv4 nf_nat iptable_mangle nf_conntrack_ipv4 nf_defrag_ipv4 xt_conntrack
nf_conntrack bnep bluetooth rfkill ebtable_filter ebtables ip6table_filter
ip6_tables snd_hda_codec_hdmi snd_cmipci snd_mpu401_uart snd_hda_intel
snd_intel8x0 snd_opl3_lib snd_ac97_codec gameport snd_hda_codec snd_rawmidi
ac97_bus snd_hwdep snd_seq snd_seq_device snd_pcm snd_page_alloc snd_timer snd
k8temp soundcore serio_raw i2c_nforce2 forcedeth ata_generic pata_acpi nouveau
video mxm_wmi wmi i2c_algo_bit drm_kms_helper ttm drm i2c_core sata_sil pata_amd
sata_nv uinput
kernel: [17539.341012] Pid: 17340, comm: sshd Not tainted
3.9.0-0.rc4.git0.1.fc19.i686.PAE #1
kernel: [17539.341012] Call Trace:
kernel: [17539.341012]  [<c045573c>] warn_slowpath_common+0x6c/0xa0
kernel: [17539.341012]  [<c0701953>] ? check_unmap+0x493/0x960
kernel: [17539.341012]  [<c0701953>] ? check_unmap+0x493/0x960
kernel: [17539.341012]  [<c04557a3>] warn_slowpath_fmt+0x33/0x40
kernel: [17539.341012]  [<c0701953>] check_unmap+0x493/0x960
kernel: [17539.341012]  [<c049238f>] ? sched_clock_cpu+0xdf/0x150
kernel: [17539.341012]  [<c0701e87>] debug_dma_unmap_page+0x67/0x70
kernel: [17539.341012]  [<f7eae8f2>] nv_unmap_txskb.isra.32+0x92/0x100

Its pretty plainly the result of an skb fragment getting unmapped without having
its initial mapping operation checked for errors.  This patch corrects that.

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: "David S. Miller" <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoISDN:divert: beautify code: useless 'break', 'return (0)', additional comments.
Chen Gang [Tue, 2 Apr 2013 01:37:56 +0000 (09:37 +0800)]
ISDN:divert: beautify code: useless 'break', 'return (0)', additional comments.

  delete useless 'break' after 'return'.
  let 'return 0' instead of 'return (0)'
  also give a comment for 'break' to let readers notice it.

Signed-off-by: Chen Gang <gang.chen@asianux.com>
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agocbq: incorrect processing of high limits
Vasily Averin [Mon, 1 Apr 2013 03:01:32 +0000 (03:01 +0000)]
cbq: incorrect processing of high limits

currently cbq works incorrectly for limits > 10% real link bandwidth,
and practically does not work for limits > 50% real link bandwidth.
Below are results of experiments taken on 1 Gbit link

 In shaper | Actual Result
-----------+---------------
  100M     | 108 Mbps
  200M     | 244 Mbps
  300M     | 412 Mbps
  500M     | 893 Mbps

This happen because of q->now changes incorrectly in cbq_dequeue():
when it is called before real end of packet transmitting,
L2T is greater than real time delay, q_now gets an extra boost
but never compensate it.

To fix this problem we prevent change of q->now until its synchronization
with real time.

Signed-off-by: Vasily Averin <vvs@openvz.org>
Reviewed-by: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: Bump up the version to 5.2.40
Rajesh Borundia [Tue, 2 Apr 2013 05:34:43 +0000 (05:34 +0000)]
qlcnic: Bump up the version to 5.2.40

Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: Fix sparse warnings.
Rajesh Borundia [Tue, 2 Apr 2013 05:34:42 +0000 (05:34 +0000)]
qlcnic: Fix sparse warnings.

warning: 'pf_info.max_tx_ques' may be used uninitialized in this function
[-Wmaybe-uninitialized]

Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: Fix NULL dereference in error path.
Rajesh Borundia [Tue, 2 Apr 2013 05:34:41 +0000 (05:34 +0000)]
qlcnic: Fix NULL dereference in error path.

o Fix for smatch tool reported error
   drivers/net/ethernet/qlogic/qlcnic/qlcnic_main.c:2029
   qlcnic_probe() error: potential NULL dereference 'adapter'.
o While returning from an error path in probe, adapter is not
  initialized. So do not access adapter in cleanup path.

Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: Fix potential NULL dereference
Rajesh Borundia [Tue, 2 Apr 2013 05:34:40 +0000 (05:34 +0000)]
qlcnic: Fix potential NULL dereference

[net-next:master 301/302] drivers/net/ethernet/qlogic/qlcnic/qlcnic_hw.c:563
 qlcnic_set_multi() error: potential null dereference 'cur'. (kzalloc returns null)

o Break out of the loop after memory allocation failure. Program all the
  MAC addresses that were cached in the return path.

Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agotcp: Remove dead sysctl_tcp_cookie_size declaration
Neal Cardwell [Tue, 2 Apr 2013 17:13:07 +0000 (13:13 -0400)]
tcp: Remove dead sysctl_tcp_cookie_size declaration

Remove a declaration left over from the TCPCT-ectomy. This sysctl is
no longer referenced anywhere since 1a2c6181c4 ("tcp: Remove TCPCT").

Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoipc: set msg back to -EAGAIN if copy wasn't performed
Stanislav Kinsbursky [Mon, 1 Apr 2013 07:40:51 +0000 (11:40 +0400)]
ipc: set msg back to -EAGAIN if copy wasn't performed

Make sure that msg pointer is set back to error value in case of
MSG_COPY flag is set and desired message to copy wasn't found.  This
garantees that msg is either a error pointer or a copy address.

Otherwise the last message in queue will be freed without unlinking from
the queue (which leads to memory corruption) and the dummy allocated
copy won't be released.

Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agobridge: remove a redundant synchronize_net()
Eric Dumazet [Tue, 2 Apr 2013 05:45:00 +0000 (05:45 +0000)]
bridge: remove a redundant synchronize_net()

commit 00cfec37484761 (net: add a synchronize_net() in
netdev_rx_handler_unregister())
allows us to remove the synchronized_net() call from del_nbp()

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet/mlx4_en: Fix setting initial MAC address
Yan Burman [Tue, 2 Apr 2013 13:49:45 +0000 (16:49 +0300)]
net/mlx4_en: Fix setting initial MAC address

Commit 6bbb6d9 "net/mlx4_en: Optimize Rx fast path filter checks" introduced a regression
under which the MAC address read from the card was not converted correctly
(the most significant byte was not handled), fix that.

Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Yan Burman <yanb@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobonding: get netdev_rx_handler_unregister out of locks
Veaceslav Falico [Tue, 2 Apr 2013 05:15:16 +0000 (05:15 +0000)]
bonding: get netdev_rx_handler_unregister out of locks

Now that netdev_rx_handler_unregister contains synchronize_net(), we need
to call it outside of bond->lock, cause it might sleep. Also, remove the
already unneded synchronize_net().

Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge tag 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
Linus Torvalds [Tue, 2 Apr 2013 15:35:03 +0000 (08:35 -0700)]
Merge tag 'fixes' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC bug fixes from Arnd Bergmann:
 "After a quiet set of fixes for 3.9-rc4, a lot of people woke up and
  sent urgent fixes for 3.9.  I pushed back on a number of them that got
  deferred to 3.10, but these are the ones that seemed important.

  Regression in 3.9:

   - Multiple regressions in OMAP2+ clock cleanup
   - SH-Mobile frame buffer bug fix that merged here because of
     maintainer MIA
   - ux500 prcmu changes broke DT booting
   - MMCI duplicated regulator setup on ux500
   - New ux500 clock driver broke ethernet on snowball
   - Local interrupt driver for mvebu broke ethernet
   - MVEBU GPIO driver did not get set up right on Orion DT
   - incorrect interrupt number on Orion crypto for DT

  Long-standing bugs, including candidates for stable:

   - Kirkwood MMC needs to disable invalid card detect pins
   - MV SDIO pinmux was wrong on Mirabox
   - GoFlex Net board file needs to set NAND chip delay
   - MSM timer restart race
   - ep93xx early debug code broke in 3.7
   - i.MX CPU hotplug race
   - Incorrect clock setup for OMAP1 USB
   - Workaround for bad clock setup by some old OMAP4 boot loaders
   - Static I/O mappings on cns3xxx since 3.2"

* tag 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  ARM: cns3xxx: fix mapping of private memory region
  arm: mvebu: Fix pinctrl for Armada 370 Mirabox SDIO port.
  arm: orion5x: correct IRQ used in dtsi for mv_cesa
  arm: orion5x: fix orion5x.dtsi gpio parameters
  ARM: Kirkwood: fix unused mvsdio gpio pins
  arm: mvebu: Use local interrupt only for the timer 0
  ARM: kirkwood: Fix chip-delay for GoFlex Net
  ARM: ux500: Enable the clock controlling Ethernet on Snowball
  ARM: ux500: Stop passing ios_handler() as an MMCI power controlling call-back
  ARM: ux500: Apply the TCPM and TCDM locations and sizes to dbx5x0 DT
  fbdev: sh_mobile_lcdc: fixup B side hsync adjust settings
  ARM: OMAP: clocks: Delay clk inits atleast until slab is initialized
  ARM: imx: fix sync issue between imx_cpu_die and imx_cpu_kill
  ARM: msm: Stop counting before reprogramming clockevent
  ARM: ep93xx: Fix wait for UART FIFO to be empty
  ARM: OMAP4: PM: fix PM regression introduced by recent clock cleanup
  ARM: OMAP3: hwmod data: keep MIDLEMODE in force-standby for musb
  ARM: OMAP4: clock data: lock USB DPLL on boot
  ARM: OMAP1: fix USB host on 1710

11 years agoMerge branch 'for-3.9' of git://linux-nfs.org/~bfields/linux
Linus Torvalds [Tue, 2 Apr 2013 14:56:20 +0000 (07:56 -0700)]
Merge branch 'for-3.9' of git://linux-nfs.org/~bfields/linux

Pull nfsd bugfix from J Bruce Fields:
 "An xdr decoding error--thanks, Toralf Förster, and Trinity!"

* 'for-3.9' of git://linux-nfs.org/~bfields/linux:
  nfsd4: reject "negative" acl lengths

11 years agoMerge tag 'v3.9-rc1_cns3xxx_fixes' of git://git.infradead.org/users/cbou/linux-cns3xx...
Arnd Bergmann [Tue, 2 Apr 2013 14:08:43 +0000 (16:08 +0200)]
Merge tag 'v3.9-rc1_cns3xxx_fixes' of git://git.infradead.org/users/cbou/linux-cns3xxx into fixes

From Anton Vorontsov <anton@enomsg.org>:

This tag includes Mac Lin's work to revive CNS3xxx booting:

 "Since commit 0536bdf33faf (ARM: move iotable mappings within the vmalloc
 region), [...] the pre-defined iotable mappings is not in the vmalloc
 region. [...] move the iotable mappings into the vmalloc region, and
 merge the MPCore private memory region (containing the SCU, the GIC and
 the TWD) as a single region."

Plus there is a small cosmetic fix, also from Mac Lin.

* tag 'v3.9-rc1_cns3xxx_fixes' of git://git.infradead.org/users/cbou/linux-cns3xxx:
  ARM: cns3xxx: fix mapping of private memory region

[arnd: dropped the cosmetic fix from the merge as it is not needed for 3.9]

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
11 years agonetfilter: fix struct ip6t_frag field description
Michal Kubeček [Tue, 2 Apr 2013 06:12:11 +0000 (08:12 +0200)]
netfilter: fix struct ip6t_frag field description

Signed-off-by: Michal Kubecek <mkubecek@suse.cz>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonetfilter: xt_NFQUEUE: coalesce IPv4 and IPv6 hashing
holger@eitzenberger.org [Sat, 23 Mar 2013 10:04:04 +0000 (10:04 +0000)]
netfilter: xt_NFQUEUE: coalesce IPv4 and IPv6 hashing

Because rev1 and rev3 of the target share the same hashing
generalize it by introduing nfqueue_hash().

Signed-off-by: Holger Eitzenberger <holger@eitzenberger.org>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agonetfilter: xt_NFQUEUE: introduce CPU fanout
holger@eitzenberger.org [Sat, 23 Mar 2013 10:04:03 +0000 (10:04 +0000)]
netfilter: xt_NFQUEUE: introduce CPU fanout

Current NFQUEUE target uses a hash, computed over source and
destination address (and other parameters), for steering the packet
to the actual NFQUEUE. This, however forgets about the fact that the
packet eventually is handled by a particular CPU on user request.

If E. g.

  1) IRQ affinity is used to handle packets on a particular CPU already
     (both single-queue or multi-queue case)

and/or

  2) RPS is used to steer packets to a specific softirq

the target easily chooses an NFQUEUE which is not handled by a process
pinned to the same CPU.

The idea is therefore to use the CPU index for determining the
NFQUEUE handling the packet.

E. g. when having a system with 4 CPUs, 4 MQ queues and 4 NFQUEUEs it
looks like this:

 +-----+  +-----+  +-----+  +-----+
 |NFQ#0|  |NFQ#1|  |NFQ#2|  |NFQ#3|
 +-----+  +-----+  +-----+  +-----+
    ^        ^        ^        ^
    |        |NFQUEUE |        |
    +        +        +        +
 +-----+  +-----+  +-----+  +-----+
 |rx-0 |  |rx-1 |  |rx-2 |  |rx-3 |
 +-----+  +-----+  +-----+  +-----+

The NFQUEUEs not necessarily have to start with number 0, setups with
less NFQUEUEs than packet-handling CPUs are not a problem as well.

This patch extends the NFQUEUE target to accept a new
NFQ_FLAG_CPU_FANOUT flag. If this is specified the target uses the
CPU index for determining the NFQUEUE being used. I have to introduce
rev3 for this. The 'flags' are folded into _v2 'bypass'.

By changing the way which queue is assigned, I'm able to improve the
performance if the processes reading on the NFQUEUs are pinned
correctly.

Signed-off-by: Holger Eitzenberger <holger@eitzenberger.org>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agoMerge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rusty...
Linus Torvalds [Mon, 1 Apr 2013 23:17:52 +0000 (16:17 -0700)]
Merge tag 'fixes-for-linus' of git://git./linux/kernel/git/rusty/linux

Pull virtio fixes from Rusty Russell:
 "One reversion, a tiny leak fix, and a cc:stable locking fix, in two
  parts"

* tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux:
  virtio: console: add locking around c_ovq operations
  virtio: console: rename cvq_lock to c_ivq_lock
  hw_random: free rng_buffer at module exit
  Revert "virtio_console: Initialize guest_connected=true for rproc_serial"

11 years agonetfilter: use IS_ENABLE to replace if defined in TRACE target
Gao feng [Thu, 21 Mar 2013 19:48:42 +0000 (19:48 +0000)]
netfilter: use IS_ENABLE to replace if defined in TRACE target

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
11 years agoloop: prevent bdev freeing while device in use
Anatol Pomozov [Mon, 1 Apr 2013 16:47:56 +0000 (09:47 -0700)]
loop: prevent bdev freeing while device in use

struct block_device lifecycle is defined by its inode (see fs/block_dev.c) -
block_device allocated first time we access /dev/loopXX and deallocated on
bdev_destroy_inode. When we create the device "losetup /dev/loopXX afile"
we want that block_device stay alive until we destroy the loop device
with "losetup -d".

But because we do not hold /dev/loopXX inode its counter goes 0, and
inode/bdev can be destroyed at any moment. Usually it happens at memory
pressure or when user drops inode cache (like in the test below). When later in
loop_clr_fd() we want to use bdev we have use-after-free error with following
stack:

BUG: unable to handle kernel NULL pointer dereference at 0000000000000280
  bd_set_size+0x10/0xa0
  loop_clr_fd+0x1f8/0x420 [loop]
  lo_ioctl+0x200/0x7e0 [loop]
  lo_compat_ioctl+0x47/0xe0 [loop]
  compat_blkdev_ioctl+0x341/0x1290
  do_filp_open+0x42/0xa0
  compat_sys_ioctl+0xc1/0xf20
  do_sys_open+0x16e/0x1d0
  sysenter_dispatch+0x7/0x1a

To prevent use-after-free we need to grab the device in loop_set_fd()
and put it later in loop_clr_fd().

The issue is reprodusible on current Linus head and v3.3. Here is the test:

  dd if=/dev/zero of=loop.file bs=1M count=1
  while [ true ]; do
    losetup /dev/loop0 loop.file
    echo 2 > /proc/sys/vm/drop_caches
    losetup -d /dev/loop0
  done

[ Doing bdgrab/bput in loop_set_fd/loop_clr_fd is safe, because every
  time we call loop_set_fd() we check that loop_device->lo_state is
  Lo_unbound and set it to Lo_bound If somebody will try to set_fd again
  it will get EBUSY.  And if we try to loop_clr_fd() on unbound loop
  device we'll get ENXIO.

  loop_set_fd/loop_clr_fd (and any other loop ioctl) is called under
  loop_device->lo_ctl_mutex. ]

Signed-off-by: Anatol Pomozov <anatol.pomozov@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipvs: do not disable bh for long time
Julian Anastasov [Fri, 22 Mar 2013 09:46:54 +0000 (11:46 +0200)]
ipvs: do not disable bh for long time

We used a global BH disable in LOCAL_OUT hook.
Add _bh suffix to all places that need it and remove
the disabling from LOCAL_OUT and sync code.

Functions like ip_defrag need protection from
BH, so add it. As for nf_nat_mangle_tcp_packet, it needs
RCU lock.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert services to rcu
Julian Anastasov [Fri, 22 Mar 2013 09:46:53 +0000 (11:46 +0200)]
ipvs: convert services to rcu

This is the final step in RCU conversion.

Things that are removed:

- svc->usecnt: now svc is accessed under RCU read lock
- svc->inc: and some unused code
- ip_vs_bind_pe and ip_vs_unbind_pe: no ability to replace PE
- __ip_vs_svc_lock: replaced with RCU
- IP_VS_WAIT_WHILE: now readers lookup svcs and dests under
RCU and work in parallel with configuration

Other changes:

- before now, a RCU read-side critical section included the
calling of the schedule method, now it is extended to include
service lookup
- ip_vs_svc_table and ip_vs_svc_fwm_table are now using hlist
- svc->pe and svc->scheduler remain to the end (of grace period),
the schedulers are prepared for such RCU readers
even after done_service is called but they need
to use synchronize_rcu because last ip_vs_scheduler_put
can happen while RCU read-side critical sections
use an outdated svc->scheduler pointer
- as planned, update_service is removed
- empty services can be freed immediately after grace period.
If dests were present, the services are freed from
the dest trash code

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert dests to rcu
Julian Anastasov [Fri, 22 Mar 2013 09:46:52 +0000 (11:46 +0200)]
ipvs: convert dests to rcu

In previous commits the schedulers started to access
svc->destinations with _rcu list traversal primitives
because the IP_VS_WAIT_WHILE macro still plays the role of
grace period. Now it is time to finish the updating part,
i.e. adding and deleting of dests with _rcu suffix before
removing the IP_VS_WAIT_WHILE in next commit.

We use the same rule for conns as for the
schedulers: dests can be searched in RCU read-side critical
section where ip_vs_dest_hold can be called by ip_vs_bind_dest.

Some things are not perfect, for example, calling
functions like ip_vs_lookup_dest from updating code under
RCU, just because we use some function both from reader
and from updater.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert sched_lock to spin lock
Julian Anastasov [Fri, 22 Mar 2013 09:46:51 +0000 (11:46 +0200)]
ipvs: convert sched_lock to spin lock

As all read_locks are gone spin lock is preferred.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: do not expect result from done_service
Julian Anastasov [Fri, 22 Mar 2013 09:46:50 +0000 (11:46 +0200)]
ipvs: do not expect result from done_service

This method releases the scheduler state,
it can not fail. Such change will help to properly
replace the scheduler in following patch.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: reorganize dest trash
Julian Anastasov [Fri, 22 Mar 2013 09:46:49 +0000 (11:46 +0200)]
ipvs: reorganize dest trash

All dests will go to trash, no exceptions.
But we have to use new list node t_list for this, due
to RCU changes in following patches. Dests will wait there
initial grace period and later all conns and schedulers to
put their reference. The dests don't get reference for
staying in dest trash as before.

As result, we do not load ip_vs_dest_put with
extra checks for last refcnt and the schedulers do not
need to play games with atomic_inc_not_zero while
selecting best destination.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert wrr scheduler to rcu
Julian Anastasov [Fri, 22 Mar 2013 09:46:48 +0000 (11:46 +0200)]
ipvs: convert wrr scheduler to rcu

The schedule method now needs _rcu list-traversal
primitive for svc->destinations. As the weight for some
dest can be reduced during dest selection, change the
algorithm to check weights by using minimum weights in the
1 .. max_weight-(di-1) range, with the same step (di). By this
way we ensure that there will be always a weight >= 1 check
before claiming that all destinations are overloaded.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert wlc scheduler to rcu
Julian Anastasov [Fri, 22 Mar 2013 09:46:47 +0000 (11:46 +0200)]
ipvs: convert wlc scheduler to rcu

The schedule method now needs _rcu list-traversal
primitive for svc->destinations.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert sh scheduler to rcu
Julian Anastasov [Fri, 22 Mar 2013 09:46:46 +0000 (11:46 +0200)]
ipvs: convert sh scheduler to rcu

Use the 3 new methods to reassign dests.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert sed scheduler to rcu
Julian Anastasov [Fri, 22 Mar 2013 09:46:45 +0000 (11:46 +0200)]
ipvs: convert sed scheduler to rcu

The schedule method now needs _rcu list-traversal
primitive for svc->destinations.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert rr scheduler to rcu
Julian Anastasov [Fri, 22 Mar 2013 09:46:44 +0000 (11:46 +0200)]
ipvs: convert rr scheduler to rcu

The schedule method now needs _rcu list-traversal
primitive for svc->destinations. As the previous entry
could be unlinked, limit the list traversals to 2 when
lookup started from previous entry.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert nq scheduler to rcu
Julian Anastasov [Fri, 22 Mar 2013 09:46:43 +0000 (11:46 +0200)]
ipvs: convert nq scheduler to rcu

The schedule method now needs _rcu list-traversal
primitive for svc->destinations.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert lc scheduler to rcu
Julian Anastasov [Fri, 22 Mar 2013 09:46:42 +0000 (11:46 +0200)]
ipvs: convert lc scheduler to rcu

The schedule method now needs _rcu list-traversal
primitive for svc->destinations.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert lblcr scheduler to rcu
Julian Anastasov [Fri, 22 Mar 2013 09:46:41 +0000 (11:46 +0200)]
ipvs: convert lblcr scheduler to rcu

The schedule method now needs _rcu list-traversal
primitive for svc->destinations. The read_lock for sched_lock is
removed. The set.lock is removed because now it is used in
rare cases, mostly under sched_lock.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert lblc scheduler to rcu
Julian Anastasov [Fri, 22 Mar 2013 09:46:40 +0000 (11:46 +0200)]
ipvs: convert lblc scheduler to rcu

The schedule method now needs _rcu list-traversal
primitive for svc->destinations. The read_lock for sched_lock is
removed. Use a dead flag to prevent new entries to be created
while scheduler is reclaimed. Use hlist for the hash table.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert dh scheduler to rcu
Julian Anastasov [Fri, 22 Mar 2013 09:46:39 +0000 (11:46 +0200)]
ipvs: convert dh scheduler to rcu

Use the new add_dest and del_dest methods
to reassign dests.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: add ip_vs_dest_hold and ip_vs_dest_put
Julian Anastasov [Fri, 22 Mar 2013 09:46:38 +0000 (11:46 +0200)]
ipvs: add ip_vs_dest_hold and ip_vs_dest_put

ip_vs_dest_hold will be used under RCU lock
while ip_vs_dest_put can be called even after dest
is removed from service, as it happens for conns and
some schedulers.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: preparations for using rcu in schedulers
Julian Anastasov [Fri, 22 Mar 2013 09:46:37 +0000 (11:46 +0200)]
ipvs: preparations for using rcu in schedulers

Allow schedulers to use rcu_dereference when
returning destination on lookup. The RCU read-side critical
section will allow ip_vs_bind_dest to get dest refcnt as
preparation for the step where destinations will be
deleted without an IP_VS_WAIT_WHILE guard that holds the
packet processing during update.

Add new optional scheduler methods add_dest,
del_dest and upd_dest. For now the methods are called
together with update_service but update_service will be
removed in a following change.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: change ip_vs_sched_lock to mutex
Julian Anastasov [Fri, 22 Mar 2013 09:46:36 +0000 (11:46 +0200)]
ipvs: change ip_vs_sched_lock to mutex

The global list with schedulers ip_vs_schedulers
is accessed only from user context - configuration and
scheduler module [un]registration. Use ip_vs_sched_mutex
instead.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: avoid kmem_cache_zalloc in ip_vs_conn_new
Julian Anastasov [Thu, 21 Mar 2013 09:58:12 +0000 (11:58 +0200)]
ipvs: avoid kmem_cache_zalloc in ip_vs_conn_new

We have many fields to set and few to reset,
use kmem_cache_alloc instead to save some cycles.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off by: Hans Schillstrom <hans@schillstrom.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: reorder keys in connection structure
Julian Anastasov [Thu, 21 Mar 2013 09:58:11 +0000 (11:58 +0200)]
ipvs: reorder keys in connection structure

__ip_vs_conn_in_get and ip_vs_conn_out_get are
hot places. Optimize them, so that ports are matched first.
By moving net and fwmark below, on 32-bit arch we can fit
caddr in 32-byte cache line and all addresses in 64-byte
cache line.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off by: Hans Schillstrom <hans@schillstrom.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert connection locking
Julian Anastasov [Thu, 21 Mar 2013 09:58:10 +0000 (11:58 +0200)]
ipvs: convert connection locking

Convert __ip_vs_conntbl_lock_array as follows:

- readers that do not modify conn lists will use RCU lock
- updaters that modify lists will use spinlock_t

Now for conn lookups we will use RCU read-side
critical section. Without using __ip_vs_conn_get such
places have access to connection fields and can
dereference some pointers like pe and pe_data plus
the ability to update timer expiration. If full access
is required we contend for reference.

We add barrier in __ip_vs_conn_put, so that
other CPUs see the refcnt operation after other writes.

With the introduction of ip_vs_conn_unlink()
we try to reorganize ip_vs_conn_expire(), so that
unhashing of connections that should stay more time is
avoided, even if it is for very short time.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off by: Hans Schillstrom <hans@schillstrom.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: convert locks used in persistence engines
Julian Anastasov [Thu, 21 Mar 2013 09:58:09 +0000 (11:58 +0200)]
ipvs: convert locks used in persistence engines

Allow the readers to use RCU lock and for
PE module registrations use global mutex instead of
spinlock. All PE modules need to use synchronize_rcu
in their module exit handler.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off by: Hans Schillstrom <hans@schillstrom.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
11 years agoipvs: remove rs_lock by using RCU
Julian Anastasov [Thu, 21 Mar 2013 09:58:08 +0000 (11:58 +0200)]
ipvs: remove rs_lock by using RCU

rs_lock was used to protect rs_table (hash table)
from updaters (under global mutex) and readers (packet handlers).
We can remove rs_lock by using RCU lock for readers. Reclaiming
dest only with kfree_rcu is enough because the readers access
only fields from the ip_vs_dest structure.

Use hlist for rs_table.

As we are now using hlist_del_rcu, introduce in_rs_table
flag as replacement for the list_empty checks which do not
work with RCU. It is needed because only NAT dests are in
the rs_table.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off by: Hans Schillstrom <hans@schillstrom.com>
Signed-off-by: Simon Horman <horms@verge.net.au>