GitHub/LineageOS/android_kernel_motorola_exynos9610.git
8 years agoopenvswitch: add mac_proto field to the flow key
Jiri Benc [Thu, 10 Nov 2016 15:28:18 +0000 (16:28 +0100)]
openvswitch: add mac_proto field to the flow key

Use a hole in the structure. We support only Ethernet so far and will add
a support for L2-less packets shortly. We could use a bool to indicate
whether the Ethernet header is present or not but the approach with the
mac_proto field is more generic and occupies the same number of bytes in the
struct, while allowing later extensibility. It also makes the code in the
next patches more self explaining.

It would be nice to use ARPHRD_ constants but those are u16 which would be
waste. Thus define our own constants.

Another upside of this is that we can overload this new field to also denote
whether the flow key is valid. This has the advantage that on
refragmentation, we don't have to reparse the packet but can rely on the
stored eth.type. This is especially important for the next patches in this
series - instead of adding another branch for L2-less packets before calling
ovs_fragment, we can just remove all those branches completely.

Signed-off-by: Jiri Benc <jbenc@redhat.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoopenvswitch: use hard_header_len instead of hardcoded ETH_HLEN
Jiri Benc [Thu, 10 Nov 2016 15:28:17 +0000 (16:28 +0100)]
openvswitch: use hard_header_len instead of hardcoded ETH_HLEN

On tx, use hard_header_len while deciding whether to refragment or drop the
packet. That way, all combinations are calculated correctly:

* L2 packet going to L2 interface (the L2 header len is subtracted),
* L2 packet going to L3 interface (the L2 header is included in the packet
  lenght),
* L3 packet going to L3 interface.

Signed-off-by: Jiri Benc <jbenc@redhat.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobpf, mlx4: fix prog refcount in mlx4_en_try_alloc_resources error path
Daniel Borkmann [Wed, 9 Nov 2016 21:02:34 +0000 (22:02 +0100)]
bpf, mlx4: fix prog refcount in mlx4_en_try_alloc_resources error path

Commit 67f8b1dcb9ee ("net/mlx4_en: Refactor the XDP forwarding rings
scheme") added a bug in that the prog's reference count is not dropped
in the error path when mlx4_en_try_alloc_resources() is failing from
mlx4_xdp_set().

We previously took bpf_prog_add(prog, priv->rx_ring_num - 1), that we
need to release again. Earlier in the call path, dev_change_xdp_fd()
itself holds a reference to the prog as well (hence the '- 1' in the
bpf_prog_add()), so a simple atomic_sub() is safe to use here. When
an error is propagated, then bpf_prog_put() is called eventually from
dev_change_xdp_fd()

Fixes: 67f8b1dcb9ee ("net/mlx4_en: Refactor the XDP forwarding rings scheme")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: ti: davinci_cpdma: free memory while channel destroy
Ivan Khoronzhuk [Tue, 8 Nov 2016 13:16:05 +0000 (15:16 +0200)]
net: ethernet: ti: davinci_cpdma: free memory while channel destroy

While create/destroy channel operation memory is not freed. It was
supposed that memory is freed while driver remove. But a channel
can be created and destroyed many times while changing number of
channels with ethtool.

Based on net-next/master

Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'hns-fixes'
David S. Miller [Thu, 10 Nov 2016 16:45:50 +0000 (11:45 -0500)]
Merge branch 'hns-fixes'

Salil Mehta says:

====================
Bug fixes & Code improvements in HNS driver

This patch-set introduces some bug fixes and code improvements.
These have been identified during internal review or testing of
the driver by internal Hisilicon teams.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: add the support to add/remove the ucast entry to/from table
Kejian Yan [Wed, 9 Nov 2016 18:14:01 +0000 (18:14 +0000)]
net: hns: add the support to add/remove the ucast entry to/from table

This patch adds the support to add or remove the unicast entries
to the table and remove from the table.

Reported-by: Daode Huang <huangdaode@hisilicon.com>
Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: add multicast tcam table clear
Kejian Yan [Wed, 9 Nov 2016 18:14:00 +0000 (18:14 +0000)]
net: hns: add multicast tcam table clear

There is no clear operation before add a new multicast tcam table,
so the tcam table will be overflow when add more entries.

Reported-by: Daode Huang <huangdaode@hisilicon.com>
Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: modify tcam table of mask_key
Qianqian Xie [Wed, 9 Nov 2016 18:13:59 +0000 (18:13 +0000)]
net: hns: modify tcam table of mask_key

The packets of wrong mac address(only the last bit is different) can be
received in Big-endian by current definition of mask_key. Thus it needs
to be modified to support Big-endian and ensure Big-endian normal.

Signed-off-by: Qianqian Xie <xieqianqian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: modify tcam table of mac mc-entry
Qianqian Xie [Wed, 9 Nov 2016 18:13:58 +0000 (18:13 +0000)]
net: hns: modify tcam table of mac mc-entry

The current definition of mac_mc_entry is only suitable for
Little-endian. Thus it needs to modify tcam table of mac mc-entry
to support both Little-endian and Big-endian.

Signed-off-by: Qianqian Xie <xieqianqian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: modify tcam table of mac mc-port
Qianqian Xie [Wed, 9 Nov 2016 18:13:57 +0000 (18:13 +0000)]
net: hns: modify tcam table of mac mc-port

Little-endian is only supported by current tcam table to add
or delete mac mc-port. This patch makes it support both
Little-endian and Big-endian.

Signed-off-by: Qianqian Xie <xieqianqian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: modify table index to get mac entry
Qianqian Xie [Wed, 9 Nov 2016 18:13:56 +0000 (18:13 +0000)]
net: hns: modify table index to get mac entry

Big-endian is not supported by the current definition of table index to get
mac entry. It needs to be modified to support both Little-endian
and Big-endian.

Signed-off-by: Qianqian Xie <xieqianqian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: modify tcam table of mac uc-entry
Qianqian Xie [Wed, 9 Nov 2016 18:13:55 +0000 (18:13 +0000)]
net: hns: modify tcam table of mac uc-entry

The current definition of mac_uc_entry is only suitable for
Little-endian. Thus it needs to modify tcam table of mac uc-entry
to support both Little-endian and Big-endian.

Signed-off-by: Qianqian Xie <xieqianqian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: modify tcam table and set mac key
Qianqian Xie [Wed, 9 Nov 2016 18:13:54 +0000 (18:13 +0000)]
net: hns: modify tcam table and set mac key

The current definition of dsaf_drv_tbl_tcam_key is only suitable for
Little-endian. If data is stored in Big-endian, this may lead to
error in data use. Shift operation can make it work normally in both
Big-endian and Little-endian.

Signed-off-by: Qianqian Xie <xieqianqian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: modify buffer format of cpu data to le64
Qianqian Xie [Wed, 9 Nov 2016 18:13:53 +0000 (18:13 +0000)]
net: hns: modify buffer format of cpu data to le64

Hardware ring buffer data is stored in Little-endian. Thus cpu data
should be modified to Little-endian.

Signed-off-by: Qianqian Xie <xieqianqian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: fix to intimate the link-status change by adding LF/RF method
Daode Huang [Wed, 9 Nov 2016 18:13:52 +0000 (18:13 +0000)]
net: hns: fix to intimate the link-status change by adding LF/RF method

In current scenario, when the interface is disabled we reset the XGMAC
RX/TX functionality. This operation does not affects the PHY layer/SFP
and which appears UP to the remote end(this behaviour is unlike GMAC).
The result is remote end keeps on sending the packets which gets partly
processed by XMAC and dropped. Since these are partly processed these
appears as errored packets in the packet counter statistics.

This patch fixes this behaviour and adds local-fault and remote-fault
functionality which can be used to intimate the remote peer whenever
the state of the interface changes. This patch also removes the
existing hns_dsaf_xge_core_srst_by_port function which was being used
to reset the RX/TX functionality at XGE Core.

Reported-by: Jun He <hjat2005@huawei.com>
Signed-off-by: Daode Huang <huangdaode@hisilicon.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: modify ethtool statistics value error
Qianqian Xie [Wed, 9 Nov 2016 18:13:51 +0000 (18:13 +0000)]
net: hns: modify ethtool statistics value error

This patch modify the gmac_rx_filt_pkt and gmac_rx_octets_total_filt
statistics value. The two statistics is inconsistent with register,
and just the opposite.

Signed-off-by: Qianqian Xie <xieqianqian@huawei.com>
Signed-off-by: Jun He <hjat2005@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: delete redundant macro definition
Qianqian Xie [Wed, 9 Nov 2016 18:13:50 +0000 (18:13 +0000)]
net: hns: delete redundant macro definition

This patch deletes redundant macro definitions in hns drivers.
And change the .h file containing relation to make the layers
more clearly

Signed-off-by: Qianqian Xie <xieqianqian@huawei.com>
Signed-off-by: Weiwei Deng <dengweiwei@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: bug fix about restart auto-negotiation
Daode Huang [Wed, 9 Nov 2016 18:13:49 +0000 (18:13 +0000)]
net: hns: bug fix about restart auto-negotiation

When set auto-negotiation off and duplex half, if run "ethtool -r ethX"
on port with phy, then the port will be failed to work. It should
forbid to start auto-negotiation when auto-negotiate is off. This
patch add the limited condition.

Reported-by: Jinchuang Tian <tianjinchuang1@huawei.com>
Signed-off-by: Daode Huang <huangdaode@hisilicon.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Reviewed-by: lipeng <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: set default mac pause time to 0xffff
Daode Huang [Wed, 9 Nov 2016 18:13:48 +0000 (18:13 +0000)]
net: hns: set default mac pause time to 0xffff

The default mac pause time set to 0xff which is too short for pausing,
this patch change it to the max value 0xffff.

Signed-off-by: Daode Huang <huangdaode@hisilicon.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Reviewed-by: lipeng <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: fix for promisc mode in HNS driver
Kejian Yan [Wed, 9 Nov 2016 18:13:47 +0000 (18:13 +0000)]
net: hns: fix for promisc mode in HNS driver

If set promisc mode when there is some traffic, The service nic will
cause system halted. We reserve the last 6 tcam entry for the 6 ports.
If promisc mode is enabled, we can config the relative tcam as fuzzy
matching and set to be valid, or set the tcam to be invalid

Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: add fuzzy match of tcam table for hns
Kejian Yan [Wed, 9 Nov 2016 18:13:46 +0000 (18:13 +0000)]
net: hns: add fuzzy match of tcam table for hns

Since there is not enough tcam table entries for vlan and multicast
address, HNSv2 needs to add support of fuzzy matching of TCAM tables.
To add fuzzy match of TCAM, we Add the property to mask the bits to
be fuzzy matched

Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoDoc: hisi: hns adds mc-mac-mask property
Kejian Yan [Wed, 9 Nov 2016 18:13:45 +0000 (18:13 +0000)]
Doc: hisi: hns adds mc-mac-mask property

Since there is not enough tcam table entries for every vlan and multicast
address, HNS needs to add support of fuzzy matching of TCAM tables. Adding
the property to mask the bits to be fuzzy matched, so update the bindings
document

Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agotcp: remove unaligned accesses from tcp_get_info()
Eric Dumazet [Wed, 9 Nov 2016 19:24:22 +0000 (11:24 -0800)]
tcp: remove unaligned accesses from tcp_get_info()

After commit 6ed46d1247a5 ("sock_diag: align nlattr properly when
needed"), tcp_get_info() gets 64bit aligned memory, so we can avoid
the unaligned helpers.

Suggested-by: David Miller <davem@davemloft.net>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'batadv-next-for-davem-20161108-v2' of git://git.open-mesh.org/linux-merge
David S. Miller [Thu, 10 Nov 2016 03:15:28 +0000 (22:15 -0500)]
Merge tag 'batadv-next-for-davem-20161108-v2' of git://git.open-mesh.org/linux-merge

Simon Wunderlich says:

====================
pull request for net-next: batman-adv 2016-11-08 v2

This feature and cleanup patchset includes the following changes:

 - netlink and code cleanups by Sven Eckelmann (3 patches)

 - Cleanup and minor fixes by Linus Luessing (3 patches)

 - Speed up multicast update intervals, by Linus Luessing

 - Avoid (re)broadcast in meshes for some easy cases,
   by Linus Luessing

 - Clean up tx return state handling, by Sven Eckelmann (6 patches)

 - Fix some special mac address handling cases, by Sven Eckelmann
   (3 patches)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'PHC-freq-fine-tuning'
David S. Miller [Thu, 10 Nov 2016 02:20:01 +0000 (21:20 -0500)]
Merge branch 'PHC-freq-fine-tuning'

Richard Cochran says:

====================
PHC frequency fine tuning

This series expands the PTP Hardware Clock subsystem by adding a
method that passes the frequency tuning word to the the drivers
without dropping the low order bits.  Keeping those bits is useful for
drivers whose frequency resolution is higher than 1 ppb.

The appended script (below) runs a simple demonstration of the
improvement.  This test needs two Intel i210 PCIe cards installed in
the same PC, with their SDP0 pins connected by copper wire.  Measuring
the estimated offset (from the ptp4l servo) and the true offset (from
the PPS) over one hour yields the following statistics.

|        |   Est. Before |    Est. After |   True Before |    True After |
|--------+---------------+---------------+---------------+---------------|
| min    | -5.200000e+01 | -1.600000e+01 | -3.100000e+01 | -1.000000e+00 |
| max    | +5.700000e+01 | +2.500000e+01 | +8.500000e+01 | +4.000000e+01 |
| pk-pk: | +1.090000e+02 | +4.100000e+01 | +1.160000e+02 | +4.100000e+01 |
| mean   | +6.472222e-02 | +1.277778e-02 | +2.422083e+01 | +1.826083e+01 |
| stddev | +1.158006e+01 | +4.581982e+00 | +1.207708e+01 | +4.981435e+00 |

Here the numbers in units of nanoseconds, and the ~20 nanosecond PPS
offset is due to input/output delays on the i210's external interface
logic.

With the series applied, both the peak to peak error and the standard
deviation improve by a factor of more than two.  These two graphs show
the improvement nicely.

  http://linuxptp.sourceforge.net/fine-tuning/fine-est.png

  http://linuxptp.sourceforge.net/fine-tuning/fine-tru.png
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoptp: dp83640: Use the high resolution frequency method.
Richard Cochran [Tue, 8 Nov 2016 21:49:18 +0000 (22:49 +0100)]
ptp: dp83640: Use the high resolution frequency method.

The dp83640 has a frequency resolution of about 0.029 ppb.
This patch lets users of the device benefit from the
increased frequency resolution when tuning the clock.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoptp: igb: Use the high resolution frequency method.
Richard Cochran [Tue, 8 Nov 2016 21:49:17 +0000 (22:49 +0100)]
ptp: igb: Use the high resolution frequency method.

The 82580 and related devices offer a frequency resolution of about
0.029 ppb.  This patch lets users of the device benefit from the
increased frequency resolution when tuning the clock.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoptp: Introduce a high resolution frequency adjustment method.
Richard Cochran [Tue, 8 Nov 2016 21:49:16 +0000 (22:49 +0100)]
ptp: Introduce a high resolution frequency adjustment method.

The internal PTP Hardware Clock (PHC) interface limits the resolution for
frequency adjustments to one part per billion.  However, some hardware
devices allow finer adjustment, and making use of the increased resolution
improves synchronization measurably on such devices.

This patch adds an alternative method that allows finer frequency tuning
by passing the scaled ppm value to PHC drivers.  This value comes from
user space, and it has a resolution of about 0.015 ppb.  We also deprecate
the older method, anticipating its removal once existing drivers have been
converted over.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Suggested-by: Ulrik De Bie <ulrik.debie-os@e2big.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: napi_hash_add() is no longer exported
Eric Dumazet [Tue, 8 Nov 2016 19:07:28 +0000 (11:07 -0800)]
net: napi_hash_add() is no longer exported

There are no more users except from net/core/dev.c
napi_hash_add() can now be static.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: do not call napi_hash_add()
Eric Dumazet [Tue, 8 Nov 2016 19:06:53 +0000 (11:06 -0800)]
bnxt_en: do not call napi_hash_add()

This is automatically done from netif_napi_add(), and we want to not
export napi_hash_add() anymore in the following patch.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Michael Chan <michael.chan@broadcom.com>
Acked-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobpf: Remove unused but set variables
Tobias Klauser [Tue, 8 Nov 2016 15:40:28 +0000 (16:40 +0100)]
bpf: Remove unused but set variables

Remove the unused but set variables min_set and max_set in
adjust_reg_min_max_vals to fix the following warning when building with
'W=1':

  kernel/bpf/verifier.c:1483:7: warning: variable â€˜min_set’ set but not used [-Wunused-but-set-variable]

There is no warning about max_set being unused, but since it is only
used in the assignment of min_set it can be removed as well.

They were introduced in commit 484611357c19 ("bpf: allow access into map
value arrays") but seem to have never been used.

Cc: Josef Bacik <jbacik@fb.com>
Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agotc_act: Remove tcf_act macro
Yotam Gigi [Tue, 8 Nov 2016 15:24:03 +0000 (17:24 +0200)]
tc_act: Remove tcf_act macro

tc_act macro addressed a non existing field, and was not used in the
kernel source.

Signed-off-by: Yotam Gigi <yotamg@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'ipv6-sr'
David S. Miller [Thu, 10 Nov 2016 01:40:13 +0000 (20:40 -0500)]
Merge branch 'ipv6-sr'

David Lebrun says:

====================
net: add support for IPv6 Segment Routing

v5:
 - Check SRH validity when adding a new route with lwtunnels and
   when setting an IPV6_RTHDR socket option.
 - Check that hdr->segments_left is not out of bounds when processing
   an SR-enabled packet.
 - Add __ro_after_init attribute to seg6_genl_policy structure.
 - Add CONFIG_IPV6_SEG6_INLINE option to enable or disable
   direct header insertion.

v4:
 - Change @cleanup in ipv6_srh_rcv() from int to bool
 - Move checksum helper functions into header file
 - Add common definition for SR TLVs
 - Add comments for HMAC computation algorithm
 - Use rhashtable to store HMAC infos instead of linked list
 - Remove packed attribute for struct sr6_tlv_hmac
 - Use dst cache only if CONFIG_DST_CACHE is enabled

v3:
 - Fix compilation for CONFIG_IPV6={n,m}

v2:
 - Remove packed attribute from sr6 struct and replaced unaligned
   16-bit flags with two 8-bit flags.
 - SR code now included by default. Option CONFIG_IPV6_SEG6_HMAC
   exists for HMAC support (which requires crypto dependencies).
 - Replace "hidden" calls to mutex_{un,}lock to direct calls.
 - Fix reverse xmas tree coding style.
 - Fix cast-from-void*'s.
 - Update skb->csum to account for SR modifications.
 - Add dst_cache in seg6_output.

Segment Routing (SR) is a source routing paradigm, architecturally
defined in draft-ietf-spring-segment-routing-09 [1]. The IPv6 flavor of
SR is defined in draft-ietf-6man-segment-routing-header-02 [2].

The main idea is that an SR-enabled packet contains a list of segments,
which represent mandatory waypoints. Each waypoint is called a segment
endpoint. The SR-enabled packet is routed normally (e.g. shortest path)
between the segment endpoints. A node that inserts an SRH into a packet
is called an ingress node, and a node that is the last segment endpoint
is called an egress node.

From an IPv6 viewpoint, an SR-enabled packet contains an IPv6 extension
header, which is a Routing Header type 4, defined as follows:

struct ipv6_sr_hdr {
        __u8    nexthdr;
        __u8    hdrlen;
        __u8    type;
        __u8    segments_left;
        __u8    first_segment;
        __u8    flag_1;
        __u8    flag_2;
        __u8    reserved;

        struct in6_addr segments[0];
};

The first 4 bytes of the SRH is consistent with the Routing Header
definition in RFC 2460. The type is set to `4' (SRH).

Each segment is encoded as an IPv6 address. The segments are encoded in
reverse order: segments[0] is the last segment of the path, and
segments[first_segment] is the first segment of the path.

segments[segments_left] points to the currently active segment and
segments_left is decremented at each segment endpoint.

There exist two ways for a packet to receive an SRH, we call them
encap mode and inline mode. In the encap mode, the packet is encapsulated
in an outer IPv6 header that contains the SRH. The inner (original) packet
is not modified. A virtual tunnel is thus created between the ingress node
(the node that encapsulates) and the egress node (the last segment of the path).
Once an encapsulated SR packet reaches the egress node, the node decapsulates
the packet and performs a routing decision on the inner packet. This kind of
SRH insertion is intended to use for routers that encapsulates in-transit
packet.

The second SRH insertion method, the inline mode, acts by directly inserting
the SRH right after the IPv6 header of the original packet. For this method,
if a particular flag (SR6_FLAG_CLEANUP) is set, then the penultimate segment
endpoint must strip the SRH from the packet before forwarding it to the last
segment endpoint. This insertion method is intended to use for endhosts,
however it is also used for in-transit packets by some industry actors.
Note that directly inserting extension headers may break several mechanisms
such as Path MTU Discovery, IPSec AH, etc. For this reason, this insertion
method is only available if CONFIG_IPV6_SEG6_INLINE is enabled.

Finally, the SRH may contain TLVs after the segments list. Several types of
TLVs are defined, but we currently consider only the HMAC TLV. This TLV is
an answer to the deprecation of the RH0 and enables to ensure the authenticity
and integrity of the SRH. The HMAC text contains the flags, the first_segment
index, the full list of segments, and the source address of the packet. While
SR is intended to use mostly within a single administrative domain, the HMAC
TLV allows to verify SR packets coming from an untrusted source.

This patches series implements support for the IPv6 flavor of SR and is
logically divided into the following components:

        (1) Data plane support (patch 01). This patch adds a function
            in net/ipv6/exthdrs.c to handle the Routing Header type 4.
            It enables the kernel to act as a segment endpoint, by supporting
            the following operations: decrementation of the segments_left field,
            cleanup flag support (removal of the SRH if we are the penultimate
            segment endpoint) and decapsulation of the inner packet as an egress
            node.

        (2) Control plane support (patches 02..03 and 07..09). These patches enables
            to insert SRH on locally emitted and/or forwarded packets, both with
            encap mode and with inline mode. The SRH insertion is controlled through
            the lightweight tunnels mechanism. Furthermore, patch 08 enables the
            applications to insert an SRH on a per-socket basis, through the
            setsockopt() system call. The mechanism to specify a per-socket
            Routing Header was already defined for RH0 and no special modification
            was performed on this side. However, the code to actually push the RH
            onto the packets had to be adapted for the SRH specifications.

        (3) HMAC support (patches 04..06). These patches adds the support of the
            HMAC TLV verification for the dataplane part, and generation for
            the control plane part. Two hashing algorithms are supported
            (SHA-1 as legacy and SHA-256 as required by the IETF draft), but
            additional algorithms can be easily supported by simply adding an
            entry into an array.

[1] https://tools.ietf.org/html/draft-ietf-spring-segment-routing-09
[2] https://tools.ietf.org/html/draft-ietf-6man-segment-routing-header-02
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: sr: add documentation file for per-interface sysctls
David Lebrun [Tue, 8 Nov 2016 13:59:22 +0000 (14:59 +0100)]
ipv6: sr: add documentation file for per-interface sysctls

This patch adds documentation for some SR-related per-interface
sysctls.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: sr: add support for SRH injection through setsockopt
David Lebrun [Tue, 8 Nov 2016 13:59:21 +0000 (14:59 +0100)]
ipv6: sr: add support for SRH injection through setsockopt

This patch adds support for per-socket SRH injection with the setsockopt
system call through the IPPROTO_IPV6, IPV6_RTHDR options.
The SRH is pushed through the ipv6_push_nfrag_opts function.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: add source address argument for ipv6_push_nfrag_opts
David Lebrun [Tue, 8 Nov 2016 13:59:20 +0000 (14:59 +0100)]
ipv6: add source address argument for ipv6_push_nfrag_opts

This patch prepares for insertion of SRH through setsockopt().
The new source address argument is used when an HMAC field is
present in the SRH, which must be filled. The HMAC signature
process requires the source address as input text.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: sr: add calls to verify and insert HMAC signatures
David Lebrun [Tue, 8 Nov 2016 13:59:19 +0000 (14:59 +0100)]
ipv6: sr: add calls to verify and insert HMAC signatures

This patch enables the verification of the HMAC signature for transiting
SR-enabled packets, and its insertion on encapsulated/injected SRH.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: sr: implement API to control SR HMAC structure
David Lebrun [Tue, 8 Nov 2016 13:59:18 +0000 (14:59 +0100)]
ipv6: sr: implement API to control SR HMAC structure

This patch provides an implementation of the genetlink commands
to associate a given HMAC key identifier with an hashing algorithm
and a secret.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: sr: add core files for SR HMAC support
David Lebrun [Tue, 8 Nov 2016 13:57:42 +0000 (14:57 +0100)]
ipv6: sr: add core files for SR HMAC support

This patch adds the necessary functions to compute and check the HMAC signature
of an SR-enabled packet. Two HMAC algorithms are supported: hmac(sha1) and
hmac(sha256).

In order to avoid dynamic memory allocation for each HMAC computation,
a per-cpu ring buffer is allocated for this purpose.

A new per-interface sysctl called seg6_require_hmac is added, allowing a
user-defined policy for processing HMAC-signed SR-enabled packets.
A value of -1 means that the HMAC field will always be ignored.
A value of 0 means that if an HMAC field is present, its validity will
be enforced (the packet is dropped is the signature is incorrect).
Finally, a value of 1 means that any SR-enabled packet that does not
contain an HMAC signature or whose signature is incorrect will be dropped.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: sr: add support for SRH encapsulation and injection with lwtunnels
David Lebrun [Tue, 8 Nov 2016 13:57:41 +0000 (14:57 +0100)]
ipv6: sr: add support for SRH encapsulation and injection with lwtunnels

This patch creates a new type of interfaceless lightweight tunnel (SEG6),
enabling the encapsulation and injection of SRH within locally emitted
packets and forwarded packets.

>From a configuration viewpoint, a seg6 tunnel would be configured as follows:

  ip -6 ro ad fc00::1/128 encap seg6 mode encap segs fc42::1,fc42::2,fc42::3 dev eth0

Any packet whose destination address is fc00::1 would thus be encapsulated
within an outer IPv6 header containing the SRH with three segments, and would
actually be routed to the first segment of the list. If `mode inline' was
specified instead of `mode encap', then the SRH would be directly inserted
after the IPv6 header without outer encapsulation.

The inline mode is only available if CONFIG_IPV6_SEG6_INLINE is enabled. This
feature was made configurable because direct header insertion may break
several mechanisms such as PMTUD or IPSec AH.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: sr: add code base for control plane support of SR-IPv6
David Lebrun [Tue, 8 Nov 2016 13:57:40 +0000 (14:57 +0100)]
ipv6: sr: add code base for control plane support of SR-IPv6

This patch adds the necessary hooks and structures to provide support
for SR-IPv6 control plane, essentially the Generic Netlink commands
that will be used for userspace control over the Segment Routing
kernel structures.

The genetlink commands provide control over two different structures:
tunnel source and HMAC data. The tunnel source is the source address
that will be used by default when encapsulating packets into an
outer IPv6 header + SRH. If the tunnel source is set to :: then an
address of the outgoing interface will be selected as the source.

The HMAC commands currently just return ENOTSUPP and will be implemented
in a future patch.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: implement dataplane support for rthdr type 4 (Segment Routing Header)
David Lebrun [Tue, 8 Nov 2016 13:57:39 +0000 (14:57 +0100)]
ipv6: implement dataplane support for rthdr type 4 (Segment Routing Header)

Implement minimal support for processing of SR-enabled packets
as described in
https://tools.ietf.org/html/draft-ietf-6man-segment-routing-header-02.

This patch implements the following operations:
- Intermediate segment endpoint: incrementation of active segment and rerouting.
- Egress for SR-encapsulated packets: decapsulation of outer IPv6 header + SRH
  and routing of inner packet.
- Cleanup flag support for SR-inlined packets: removal of SRH if we are the
  penultimate segment endpoint.

A per-interface sysctl seg6_enabled is provided, to accept/deny SR-enabled
packets. Default is deny.

This patch does not provide support for HMAC-signed packets.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: mii: report 0 for unknown lp_advertising
Arnd Bergmann [Tue, 8 Nov 2016 13:31:38 +0000 (14:31 +0100)]
net: mii: report 0 for unknown lp_advertising

The newly introduced mii_ethtool_get_link_ksettings function sets
lp_advertising to an uninitialized value when BMCR_ANENABLE is not
set:

drivers/net/mii.c: In function 'mii_ethtool_get_link_ksettings':
drivers/net/mii.c:224:2: error: 'lp_advertising' may be used uninitialized in this function [-Werror=maybe-uninitialized]

As documented in include/uapi/linux/ethtool.h, the value is
expected to be zero when we don't know it, so let's initialize
it to that.

Fixes: bc8ee596afe8 ("net: mii: add generic function to support ksetting support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoxen-netback: prefer xenbus_scanf() over xenbus_gather()
Jan Beulich [Tue, 8 Nov 2016 07:45:53 +0000 (00:45 -0700)]
xen-netback: prefer xenbus_scanf() over xenbus_gather()

For single items being collected this should be preferred as being more
typesafe (as the compiler can check format string and to-be-written-to
variable match) and more efficient (requiring one less parameter to be
passed).

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Paul Durrant <paul.durrant@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoigmp: Document sysctl force_igmp_version
Hangbin Liu [Mon, 7 Nov 2016 06:51:23 +0000 (14:51 +0800)]
igmp: Document sysctl force_igmp_version

There is some difference between force_igmp_version and force_mld_version.
Add document to make users aware of this.

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: l2tp: fix negative assignment to unsigned int
Asbjørn Sloth Tønnesen [Mon, 7 Nov 2016 20:39:28 +0000 (20:39 +0000)]
net: l2tp: fix negative assignment to unsigned int

recv_seq, send_seq and lns_mode mode are all defined as
unsigned int foo:1;

Signed-off-by: Asbjoern Sloth Toennesen <asbjorn@asbjorn.st>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: l2tp: cleanup: remove redundant condition
Asbjørn Sloth Tønnesen [Mon, 7 Nov 2016 20:39:27 +0000 (20:39 +0000)]
net: l2tp: cleanup: remove redundant condition

These assignments follow this pattern:

unsigned int foo:1;
struct nlattr *nla = info->attrs[bar];

if (nla)
foo = nla_get_flag(nla); /* expands to: foo = !!nla */

This could be simplified to: if (nla) foo = 1;
but lets just remove the condition and use the macro,

foo = nla_get_flag(nla);

Signed-off-by: Asbjoern Sloth Toennesen <asbjorn@asbjorn.st>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: l2tp: netlink: l2tp_nl_tunnel_send: set UDP6 checksum flags
Asbjørn Sloth Tønnesen [Mon, 7 Nov 2016 20:39:26 +0000 (20:39 +0000)]
net: l2tp: netlink: l2tp_nl_tunnel_send: set UDP6 checksum flags

This patch causes the proper attribute flags to be set,
in the case that IPv6 UDP checksums are disabled, so that
userspace ie. `ip l2tp show tunnel` knows about it.

Signed-off-by: Asbjoern Sloth Toennesen <asbjorn@asbjorn.st>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: l2tp: only set L2TP_ATTR_UDP_CSUM if AF_INET
Asbjørn Sloth Tønnesen [Mon, 7 Nov 2016 20:39:25 +0000 (20:39 +0000)]
net: l2tp: only set L2TP_ATTR_UDP_CSUM if AF_INET

Only set L2TP_ATTR_UDP_CSUM in l2tp_nl_tunnel_send()
when it's running over IPv4.

This prepares the code to also have IPv6 specific attributes.

Signed-off-by: Asbjoern Sloth Toennesen <asbjorn@asbjorn.st>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: l2tp: change L2TP_ATTR_UDP_ZERO_CSUM6_{RX, TX} attribute types
Asbjørn Sloth Tønnesen [Mon, 7 Nov 2016 20:39:24 +0000 (20:39 +0000)]
net: l2tp: change L2TP_ATTR_UDP_ZERO_CSUM6_{RX, TX} attribute types

The attributes L2TP_ATTR_UDP_ZERO_CSUM6_RX and
L2TP_ATTR_UDP_ZERO_CSUM6_TX are used as flags,
but is defined as a u8 in a comment.

This patch redocuments them as flags.

Adding nla_policy entries would break API, so not doing that.

CC: Tom Herbert <therbert@google.com>
Signed-off-by: Asbjoern Sloth Toennesen <asbjorn@asbjorn.st>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet-gro: avoid reorders
Eric Dumazet [Mon, 7 Nov 2016 19:12:27 +0000 (11:12 -0800)]
net-gro: avoid reorders

Receiving a GSO packet in dev_gro_receive() is not uncommon
in stacked devices, or devices partially implementing LRO/GRO
like bnx2x. GRO is implementing the aggregation the device
was not able to do itself.

Current code causes reorders, like in following case :

For a given flow where sender sent 3 packets P1,P2,P3,P4

Receiver might receive P1 as a single packet, stored in GRO engine.

Then P2-P4 are received as a single GSO packet, immediately given to
upper stack, while P1 is held in GRO engine.

This patch will make sure P1 is given to upper stack, then P2-P4
immediately after.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'sfc-udp-rss'
David S. Miller [Wed, 9 Nov 2016 18:59:17 +0000 (13:59 -0500)]
Merge branch 'sfc-udp-rss'

Edward Cree says:

====================
sfc: enable 4-tuple UDP RSS hashing

EF10 based NICs have configurable RSS hash fields, and can be made to take the
ports into the hash on UDP (they already do so for TCP).  This patch series
enables this, in order to improve spreading of UDP traffic.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agosfc: report 4-tuple UDP hashing to ethtool, if it's enabled
Edward Cree [Thu, 3 Nov 2016 22:12:58 +0000 (22:12 +0000)]
sfc: report 4-tuple UDP hashing to ethtool, if it's enabled

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agosfc: enable 4-tuple RSS hashing for UDP
Edward Cree [Thu, 3 Nov 2016 22:12:27 +0000 (22:12 +0000)]
sfc: enable 4-tuple RSS hashing for UDP

This improves UDP spreading, and also slightly improves GRO performance
of encapsulated TCP on 7000 series NICs.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'mlx5-SRIOV-offload-tunnel_key-set-release'
David S. Miller [Wed, 9 Nov 2016 18:41:57 +0000 (13:41 -0500)]
Merge branch 'mlx5-SRIOV-offload-tunnel_key-set-release'

Saeed Mahameed says:

====================
Mellanox 100G SRIOV offloads tunnel_key set/release

From Hadar Hen Zion:

This series further enhances the SRIOV TC offloads of mlx5 to handle the
TC tunnel_key release and set actions.

This serves a common use-case in virtualization systems where the virtual
switch encapsulate packets (tunnel_key set action) sent from VMs with
outer headers corresponding to the local/remote host IPs and de-capsulate
(tunnel_key release) outer headers before the packets are received by the
VM.

We use the new E-Switch switchdev mode and TC tunnel_key set/release
action to achieve that also in SW defined SRIOV environments by
offloading TC rules that contain these actions along with forwarding
(TC mirred/redirect action) the packets.

The first six patches are adding the needed support in flow dissector,
flower and tc for offloading tunnel_key actions:
    - The first three patches are adding the needed help functions
      and enums
    - The next three patches in the series are adding UDP port attribute
      to tunnel_key release and set actions.

The addition of UDP ports would allow the HW driver to make sure they are
given (say) a VXLAN tunnel to offload (mlx5e uses that).

Patches 7-10 are mlx5 preparations for tunnel_key actions offloads support.

Patch #11 adds mlx5e support to offload tunnel_key release action, and the
last two patches (#12-13) add mlx5e support to tc tunnel_key set action.

Currently in order to offload tc tunnel_key release action, the tc rule
should be placed on top of the mlx5e offloading (uplink) interface instead
of the shared tunnel interface. The resolution between the tunnel interface
to the HW netdevice will be implemented in a follow up series.

This series was generated against commit
94edc86bf13f ("Merge branch 'dwmac-sti-refactor-cleanup'")
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5e: Add basic TC tunnel set action for SRIOV offloads
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:48 +0000 (15:14 +0200)]
net/mlx5e: Add basic TC tunnel set action for SRIOV offloads

In mlx5 HW, encapsulation is offloaded by the steering rule having
index into an encapsulation table containing the entire set of headers
to be added by the HW. The driver sets these headers in a buffer when we
are offloading the action.

The code maintains mlx5_encap_entry for each encap header it has
encountered when attempted to offload TC tunnel set action.

This entry maintains a linked list of all the flows sharing the same
encap header, when the last flow is removed from the list the encap
entry is removed.

The actual encap_header is allocated by the driver in the hardware only
if we have layer two neighbour info when the encap entry is created.
While the flow is in the driver, the driver holds a reference on the
neighbour.

When a new flow with encap action is inserted, the code first checks if
the required encap entry exists according to the tunnel set parameters.
If it does the encap is shared, otherwise a new mlx5_encap_entry is
created.

TC action parsing implementation in the driver assumes that tunnel set
action is provided in the same order set by the user, e.g before the
mirred_redirect action.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5e: Add ndo_udp_tunnel_add to VF representors
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:47 +0000 (15:14 +0200)]
net/mlx5e: Add ndo_udp_tunnel_add to VF representors

By implementing this ndo, the host stack will set the vxlan udp port
also to VF representor netdevices. This will allow the TC offload code
in the driver when it gets a tunnel key set action to identify the UDP
port as vxlan, and hence the rule will be a candidate for offloading.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5e: Add TC tunnel release action for SRIOV offloads
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:46 +0000 (15:14 +0200)]
net/mlx5e: Add TC tunnel release action for SRIOV offloads

Enhance the parsing of offloaded TC rules to set HW matching on outer
(encapsulation) headers.
Parse TC tunnel release action and set it as mlx5 decap action when the
required capabilities are supported.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5: Support encap id when setting new steering entry
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:45 +0000 (15:14 +0200)]
net/mlx5: Support encap id when setting new steering entry

In order to support steering rules which add encapsulation headers,
encap_id parameter is needed.

Add new mlx5_flow_act struct which holds action related parameter:
action, flow_tag and encap_id. Use mlx5_flow_act struct when adding a new
steering rule.
This patch doesn't change any functionality.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5: Add creation flags when adding new flow table
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:44 +0000 (15:14 +0200)]
net/mlx5: Add creation flags when adding new flow table

When creating flow tables, allow the caller to specify creation flags.
Currently no flags are used and as such this patch doesn't add any new
functionality.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5: Check max encap header size capability
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:43 +0000 (15:14 +0200)]
net/mlx5: Check max encap header size capability

Instead of comparing to a const value, check the value of max encap
header size capability as reported by the Firmware.

Fixes: 575ddf5888ea ('net/mlx5: Introduce alloc_encap and dealloc_encap commands')
Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5: Move alloc/dealloc encap commands declarations to common header file
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:42 +0000 (15:14 +0200)]
net/mlx5: Move alloc/dealloc encap commands declarations to common header file

The alloc and dealloc encap commands will be used in the mlx5e driver,
as such, declare them in a common header file.

Also, rename the functions: mlx5_cmd_{de}alloc_encap is replaced with
mlx5_encap_{de}alloc.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/sched: act_tunnel_key: Add UDP dst port option
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:41 +0000 (15:14 +0200)]
net/sched: act_tunnel_key: Add UDP dst port option

The current tunnel set action supports only IP addresses and key
options. Add UDP dst port option.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/dst: Add dst port to dst_metadata utility functions
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:40 +0000 (15:14 +0200)]
net/dst: Add dst port to dst_metadata utility functions

Add dst port parameter to __ip_tun_set_dst and __ipv6_tun_set_dst
utility functions.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/sched: cls_flower: Add UDP port to tunnel parameters
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:39 +0000 (15:14 +0200)]
net/sched: cls_flower: Add UDP port to tunnel parameters

The current IP tunneling classification supports only IP addresses and key.
Enhance UDP based IP tunneling classification parameters by adding UDP
src and dst port.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/sched: cls_flower: Allow setting encapsulation fields as used key
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:38 +0000 (15:14 +0200)]
net/sched: cls_flower: Allow setting encapsulation fields as used key

When encapsulation field is set, mark it as used key for the flow
dissector. This will be used by offloading drivers.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoflow_dissector: Add enums for encapsulation keys
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:37 +0000 (15:14 +0200)]
flow_dissector: Add enums for encapsulation keys

New encapsulation keys were added to the flower classifier, which allow
classification according to outer (encapsulation) headers attributes
such as key and IP addresses.
In order to expose those attributes outside flower, add
corresponding enums in the flow dissector.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/sched: act_tunnel_key: add helper inlines to access tcf_tunnel_key
Hadar Hen Zion [Mon, 7 Nov 2016 13:14:36 +0000 (15:14 +0200)]
net/sched: act_tunnel_key: add helper inlines to access tcf_tunnel_key

Needed for drivers to pick the relevant action when offloading tunnel
key act.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: core: add missing check for uid_range in rule_exists.
Lorenzo Colitti [Sun, 6 Nov 2016 15:16:25 +0000 (00:16 +0900)]
net: core: add missing check for uid_range in rule_exists.

Without this check, it is not possible to create two rules that
are identical except for their UID ranges. For example:

root@net-test:/# ip rule add prio 1000 lookup 300
root@net-test:/# ip rule add prio 1000 uidrange 100-200 lookup 300
RTNETLINK answers: File exists
root@net-test:/# ip rule add prio 1000 uidrange 100-199 lookup 100
root@net-test:/# ip rule add prio 1000 uidrange 200-299 lookup 200
root@net-test:/# ip rule add prio 1000 uidrange 300-399 lookup 100
RTNETLINK answers: File exists

Tested: https://android-review.googlesource.com/#/c/299980/
Signed-off-by: Lorenzo Colitti <lorenzo@google.com>
Acked-by: Maciej Żenczykowski <maze@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqed: Prevent stack corruption on MFW interaction
Mintz, Yuval [Sun, 6 Nov 2016 15:12:27 +0000 (17:12 +0200)]
qed: Prevent stack corruption on MFW interaction

Driver uses a union for copying data to & from management firmware
when interacting with it.
Problem is that the function always copies sizeof(union) while commit
2edbff8dcb5d ("qed: Learn resources from management firmware") is casting
a union elements which is of smaller size [24-byte instead of 88-bytes].

Also, the union contains some inappropriate elements which increase its
size [should have been 32-bytes]. While this shouldn't corrupt other
PF messages to the MFW [as management firmware enforces permissions so
that each PF is allowed to write only to its own mailbox] we fix this
here as well.

Fixes: 2edbff8dcb5d ("qed: Learn resources from management firmware")
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: 3com: typhoon: fix typhoon_get_link_ksettings
Philippe Reynes [Sun, 6 Nov 2016 14:02:32 +0000 (15:02 +0100)]
net: 3com: typhoon: fix typhoon_get_link_ksettings

When moving from typhoon_get_settings to typhoon_getlink_ksettings
in the commit f7a5537cd2a5 ("net: 3com: typhoon: use new api
ethtool_{get|set}_link_ksettings"), we use a local variable supported
but we forgot to update the struct ethtool_link_ksettings with
this value.

We also initialize advertising to zero, because otherwise it may
be uninitialized if no case of the switch (tp->xcvr_select) is used.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: xgbe: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Sun, 6 Nov 2016 13:57:04 +0000 (14:57 +0100)]
net: xgbe: use new api ethtool_{get|set}_link_ksettings

The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Acked-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: amd: pcnet32: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Sat, 5 Nov 2016 23:26:41 +0000 (00:26 +0100)]
net: amd: pcnet32: use new api ethtool_{get|set}_link_ksettings

The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: amd8111e: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Sat, 5 Nov 2016 19:17:03 +0000 (20:17 +0100)]
net: amd8111e: use new api ethtool_{get|set}_link_ksettings

The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: alteon: acenic: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Sat, 5 Nov 2016 15:17:54 +0000 (16:17 +0100)]
net: alteon: acenic: use new api ethtool_{get|set}_link_ksettings

The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Acked-by: Jes Sorensen <Jes.Sorensen@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: adaptec: starfire: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Sat, 5 Nov 2016 13:05:39 +0000 (14:05 +0100)]
net: adaptec: starfire: use new api ethtool_{get|set}_link_ksettings

The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'stmmac-dwmac-rk-PM'
David S. Miller [Wed, 9 Nov 2016 18:21:25 +0000 (13:21 -0500)]
Merge branch 'stmmac-dwmac-rk-PM'

Joachim Eastwood says:

====================
stmmac: dwmac-rk: convert to standard PM/remove functions

This patch set aims to remove the init/exit callbacks from the
dwmac-rk driver and instead use standard PM callbacks. Eventually
the init/exit callbacks will be deprecated and removed from all
drivers dwmac-* except for dwmac-generic. Drivers will be refactored
to use standard PM and remove callbacks.

This conversion was pretty straight forward, but it would really nice
if some chromium people could test suspend/resume with this patch set.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoRevert "net: stmmac: allow to split suspend/resume from init/exit callbacks"
Joachim Eastwood [Sat, 5 Nov 2016 13:04:52 +0000 (14:04 +0100)]
Revert "net: stmmac: allow to split suspend/resume from init/exit callbacks"

Instead of adding hooks inside stmmac_platform it is better to just use
the standard PM callbacks within the specific dwmac-driver. This only
used by the dwmac-rk driver.

This reverts commit cecbc5563a02 ("stmmac: allow to split suspend/resume
from init/exit callbacks").

Signed-off-by: Joachim Eastwood <manabian@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agostmmac: dwmac-rk: absorb rk_gmac_init into probe
Joachim Eastwood [Sat, 5 Nov 2016 13:04:51 +0000 (14:04 +0100)]
stmmac: dwmac-rk: absorb rk_gmac_init into probe

Since the rk_gmac_init() only calls another function move this
function call into probe so rk_gmac_init() can be removed.

Since commit cecbc5563a02 ("stmmac: allow to split suspend/resume
from init/exit callbacks") the init hook is no longer used in
dwmac-rk so this can be removed.

Signed-off-by: Joachim Eastwood <manabian@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agostmmac: dwmac-rk: turn exit into standard driver remove callback
Joachim Eastwood [Sat, 5 Nov 2016 13:04:50 +0000 (14:04 +0100)]
stmmac: dwmac-rk: turn exit into standard driver remove callback

Convert the exit hook into a standard driver remove function as
the hook doesn't really buy us anything extra.

Eventually the exit hook will be deprecated in favor of the driver
remove function.

Signed-off-by: Joachim Eastwood <manabian@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agostmmac: dwmac-rk: turn resume/suspend into standard PM callbacks
Joachim Eastwood [Sat, 5 Nov 2016 13:04:49 +0000 (14:04 +0100)]
stmmac: dwmac-rk: turn resume/suspend into standard PM callbacks

Use standard PM resume/suspend callbacks instead of the hooks in
stmmac_platform. This gives the driver more control and flexibility
when implementing PM functionality. The hooks in stmmac_platform
also doesn't buy us anything extra.

Signed-off-by: Joachim Eastwood <manabian@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'tcp_get_info-locking'
David S. Miller [Wed, 9 Nov 2016 18:02:28 +0000 (13:02 -0500)]
Merge branch 'tcp_get_info-locking'

Eric Dumazet says:

====================
tcp: tcp_get_info() locking changes

This short series prepares tcp_get_info() for more detailed infos.

In order to not slow down fast path, our goal is to use the normal
socket spinlock instead of custom synchronization.

All we need to ensure is that tcp_get_info() is not called with
ehash lock, which might dead lock, since packet processing would acquire
the spinlocks in reverse way.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agotcp: no longer hold ehash lock while calling tcp_get_info()
Eric Dumazet [Fri, 4 Nov 2016 18:54:32 +0000 (11:54 -0700)]
tcp: no longer hold ehash lock while calling tcp_get_info()

We had various problems in the past in tcp_get_info() and used
specific synchronization to avoid deadlocks.

We would like to add more instrumentation points for TCP, and
avoiding grabing socket lock in tcp_getinfo() was too costly.

Being able to lock the socket allows to provide consistent set
of fields.

inet_diag_dump_icsk() can make sure ehash locks are not
held any more when tcp_get_info() is called.

We can remove syncp added in commit d654976cbf85
("tcp: fix a potential deadlock in tcp_get_info()"), but we need
to use lock_sock_fast() instead of spin_lock_bh() since TCP input
path can now be run from process context.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agotcp: shortcut listeners in tcp_get_info()
Eric Dumazet [Fri, 4 Nov 2016 18:54:31 +0000 (11:54 -0700)]
tcp: shortcut listeners in tcp_get_info()

Being lockless in tcp_get_info() is hard, because we need to add
specific synchronization in TCP fast path, like seqcount.

Following patch will change inet_diag_dump_icsk() to no longer
hold any lock for non listeners, so that we can properly acquire
socket lock in get_tcp_info() and let it return more consistent counters.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'Meson-GXL-internal-phy'
David S. Miller [Wed, 9 Nov 2016 17:50:56 +0000 (12:50 -0500)]
Merge branch 'Meson-GXL-internal-phy'

Neil Armstrong says:

====================
ARM64: Add Internal PHY support for Meson GXL

The Amlogic Meson GXL SoCs have an internal RMII PHY that is muxed with the
external RGMII pins.

In order to support switching between the two PHYs links, extended registers
size for mdio-mux-mmioreg must be added.

The DT related patches submitted as RFC in [3] will be sent in a separate
patchset due to multiple patchsets and DTSI migrations.

Changes since v2 RFC patchset at : [3]
 - Change phy Kconfig/Makefile alphabetic order
 - GXL dtsi cleanup

Changes since original RFC patchset at : [2]
 - Remove meson8b experimental phy switching
 - Switch to mdio-mux-mmioreg with extennded size support
 - Add internal phy support for S905x and p231
 - Add external PHY support for p230

[1] http://lkml.kernel.org/r/1477932286-27482-1-git-send-email-narmstrong@baylibre.com
[2] http://lkml.kernel.org/r/1477060838-14164-1-git-send-email-narmstrong@baylibre.com
[3] http://lkml.kernel.org/r/1477932987-27871-1-git-send-email-narmstrong@baylibre.com
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: phy: Add Meson GXL Internal PHY driver
Neil Armstrong [Fri, 4 Nov 2016 15:51:23 +0000 (16:51 +0100)]
net: phy: Add Meson GXL Internal PHY driver

Add driver for the Internal RMII PHY found in the Amlogic Meson GXL SoCs.

This PHY seems to only implement some standard registers and need some
workarounds to provide autoneg values from vendor registers.

Some magic values are currently used to configure the PHY, and this a
temporary setup until clarification about these registers names and
registers fields are provided by Amlogic.

Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: mdio-mux-mmioreg: Add support for 16bit and 32bit register sizes
Neil Armstrong [Fri, 4 Nov 2016 15:51:22 +0000 (16:51 +0100)]
net: mdio-mux-mmioreg: Add support for 16bit and 32bit register sizes

In order to support PHY switching on Amlogic GXL SoCs, add support for
16bit and 32bit registers sizes.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'rds-tcp-fixes'
David S. Miller [Wed, 9 Nov 2016 17:47:50 +0000 (12:47 -0500)]
Merge branch 'rds-tcp-fixes'

Sowmini Varadhan says:

====================
RDS: TCP: bug fixes

A couple of bug fixes identified during testing.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoRDS: TCP: start multipath acceptor loop at 0
Sowmini Varadhan [Fri, 4 Nov 2016 17:04:12 +0000 (10:04 -0700)]
RDS: TCP: start multipath acceptor loop at 0

The for() loop in rds_tcp_accept_one() assumes that the 0'th
rds_tcp_conn_path is UP and starts multipath accepts at index 1.
But this assumption may not always be true: if the 0'th path
has failed (ERROR or DOWN state) an incoming connection request
should be used to resurrect this path.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoRDS: TCP: report addr/port info based on TCP socket in rds-info
Sowmini Varadhan [Fri, 4 Nov 2016 17:04:11 +0000 (10:04 -0700)]
RDS: TCP: report addr/port info based on TCP socket in rds-info

The socket argument passed to rds_tcp_tc_info() is a PF_RDS socket,
so it is incorrect to report the address port info based on
rds_getname() as part of TCP state report.

Invoke inet_getname() for the t_sock associated with the
rds_tcp_connection instead.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobatman-adv: Reject unicast packet with zero/mcast dst address
Sven Eckelmann [Sat, 6 Aug 2016 15:04:23 +0000 (17:04 +0200)]
batman-adv: Reject unicast packet with zero/mcast dst address

An unicast batman-adv packet cannot be transmitted to a multicast or zero
mac address. So reject incoming packets which still have these classes of
addresses as destination mac address in the outer ethernet header.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
8 years agobatman-adv: Disallow zero and mcast src address for mgmt frames
Sven Eckelmann [Sat, 6 Aug 2016 15:04:22 +0000 (17:04 +0200)]
batman-adv: Disallow zero and mcast src address for mgmt frames

The routing check for management frames is validating the source mac
address in the outer ethernet header. It rejects every source mac address
which is a broadcast address. But it also has to reject the zero-mac
address and multicast mac addresses.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
8 years agobatman-adv: Disallow mcast src address for data frames
Sven Eckelmann [Sat, 6 Aug 2016 15:04:21 +0000 (17:04 +0200)]
batman-adv: Disallow mcast src address for data frames

The routing checks are validating the source mac address of the outer
ethernet header. They reject every source mac address which is a broadcast
address. But they also have to reject any multicast mac addresses.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
[sw@simonwunderlich.de: fix commit message typo]
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
8 years agobatman-adv: Remove dev_queue_xmit return code exception
Sven Eckelmann [Sun, 17 Jul 2016 19:04:05 +0000 (21:04 +0200)]
batman-adv: Remove dev_queue_xmit return code exception

No caller of batadv_send_skb_to_orig is expecting the results to be -1
(-EPERM) anymore when the skbuff was not consumed. They will instead expect
that the skbuff is always consumed. Having such return code filter is
therefore not needed anymore.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
8 years agobatman-adv: Consume skb in receive handlers
Sven Eckelmann [Sun, 17 Jul 2016 19:04:04 +0000 (21:04 +0200)]
batman-adv: Consume skb in receive handlers

Receiving functions in Linux consume the supplied skbuff. Doing the same in
the batadv_rx_handler functions makes the behavior more similar to the rest
of the Linux network code.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
8 years agosock: do not set sk_err in sock_dequeue_err_skb
Soheil Hassas Yeganeh [Thu, 3 Nov 2016 22:24:27 +0000 (18:24 -0400)]
sock: do not set sk_err in sock_dequeue_err_skb

Do not set sk_err when dequeuing errors from the error queue.
Doing so results in:
a) Bugs: By overwriting existing sk_err values, it possibly
   hides legitimate errors. It is also incorrect when local
   errors are queued with ip_local_error. That happens in the
   context of a system call, which already returns the error
   code.
b) Inconsistent behavior: When there are pending errors on
   the error queue, sk_err is sometimes 0 (e.g., for
   the first timestamp on the error queue) and sometimes
   set to an error code (after dequeuing the first
   timestamp).
c) Suboptimality: Setting sk_err to ENOMSG on simple
   TX timestamps can abort parallel reads and writes.

Removing this line doesn't break userspace. This is because
userspace code cannot rely on sk_err for detecting whether
there is something on the error queue. Except for ICMP messages
received for UDP and RAW, sk_err is not set at enqueue time,
and as a result sk_err can be 0 while there are plenty of
errors on the error queue.

For ICMP packets in UDP and RAW, sk_err is set when they are
enqueued on the error queue, but that does not result in aborting
reads and writes. For such cases, sk_err is only readable via
getsockopt(SO_ERROR) which will reset the value of sk_err on
its own. More importantly, prior to this patch,
recvmsg(MSG_ERRQUEUE) has a race on setting sk_err (i.e.,
sk_err is set by sock_dequeue_err_skb without atomic ops or
locks) which can store 0 in sk_err even when we have ICMP
messages pending. Removing this line from sock_dequeue_err_skb
eliminates that race.

Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'IFF_NO_QUEUE-semantics'
David S. Miller [Tue, 8 Nov 2016 01:15:56 +0000 (20:15 -0500)]
Merge branch 'IFF_NO_QUEUE-semantics'

Jesper Dangaard Brouer says:

====================
qdisc and tx_queue_len cleanups for IFF_NO_QUEUE devices

This patchset is a cleanup for IFF_NO_QUEUE devices.  It will
hopefully help userspace get a more consistent behavior when attaching
qdisc to such virtual devices.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqdisc: catch misconfig of attaching qdisc to tx_queue_len zero device
Jesper Dangaard Brouer [Thu, 3 Nov 2016 13:56:11 +0000 (14:56 +0100)]
qdisc: catch misconfig of attaching qdisc to tx_queue_len zero device

It is a clear misconfiguration to attach a qdisc to a device with
tx_queue_len zero, because some qdisc's (namely, pfifo, bfifo, gred,
htb, plug and sfb) inherit/copy this value as their queue length.

Why should the kernel catch such a misconfiguration?  Because prior to
introducing the IFF_NO_QUEUE device flag, userspace found a loophole
in the qdisc config system that allowed them to achieve the equivalent
of IFF_NO_QUEUE, which is to remove the qdisc code path entirely from
a device.  The loophole on older kernels is setting tx_queue_len=0,
*prior* to device qdisc init (the config time is significant, simply
setting tx_queue_len=0 doesn't trigger the loophole).

This loophole is currently used by Docker[1] to get better performance
and scalability out of the veth device.  The Docker developers were
warned[1] that they needed to adjust the tx_queue_len if ever
attaching a qdisc.  The OpenShift project didn't remember this warning
and attached a qdisc, this were caught and fixed in[2].

[1] https://github.com/docker/libcontainer/pull/193
[2] https://github.com/openshift/origin/pull/11126

Instead of fixing every userspace program that used this loophole, and
forgot to reset the tx_queue_len, prior to attaching a qdisc.  Let's
catch the misconfiguration on the kernel side.

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/qdisc: IFF_NO_QUEUE drivers should use consistent TX queue len
Jesper Dangaard Brouer [Thu, 3 Nov 2016 13:56:06 +0000 (14:56 +0100)]
net/qdisc: IFF_NO_QUEUE drivers should use consistent TX queue len

The flag IFF_NO_QUEUE marks virtual device drivers that doesn't need a
default qdisc attached, given they will be backed by physical device,
that already have a qdisc attached for pushback.

It is still supported to attach a qdisc to a IFF_NO_QUEUE device, as
this can be useful for difference policy reasons (e.g. bandwidth
limiting containers).  For this to work, the tx_queue_len need to have
a sane value, because some qdiscs inherit/copy the tx_queue_len
(namely, pfifo, bfifo, gred, htb, plug and sfb).

Commit a813104d9233 ("IFF_NO_QUEUE: Fix for drivers not calling
ether_setup()") caught situations where some drivers didn't initialize
tx_queue_len.  The problem with the commit was choosing 1 as the
fallback value.

A qdisc queue length of 1 causes more harm than good, because it
creates hard to debug situations for userspace. It gives userspace a
false sense of a working config after attaching a qdisc.  As low
volume traffic (that doesn't activate the qdisc policy) works,
like ping, while traffic that e.g. needs shaping cannot reach the
configured policy levels, given the queue length is too small.

This patch change the value to DEFAULT_TX_QUEUE_LEN, given other
IFF_NO_QUEUE devices (that call ether_setup()) also use this value.

Fixes: a813104d9233 ("IFF_NO_QUEUE: Fix for drivers not calling ether_setup()")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: make default TX queue length a defined constant
Jesper Dangaard Brouer [Thu, 3 Nov 2016 13:56:01 +0000 (14:56 +0100)]
net: make default TX queue length a defined constant

The default TX queue length of Ethernet devices have been a magic
constant of 1000, ever since the initial git import.

Looking back in historical trees[1][2] the value used to be 100,
with the same comment "Ethernet wants good queues". The commit[3]
that changed this from 100 to 1000 didn't describe why, but from
conversations with Robert Olsson it seems that it was changed
when Ethernet devices went from 100Mbit/s to 1Gbit/s, because the
link speed increased x10 the queue size were also adjusted.  This
value later caused much heartache for the bufferbloat community.

This patch merely moves the value into a defined constant.

[1] https://git.kernel.org/cgit/linux/kernel/git/davem/netdev-vger-cvs.git/
[2] https://git.kernel.org/cgit/linux/kernel/git/tglx/history.git/
[3] https://git.kernel.org/tglx/history/c/98921832c232

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>