GitHub/moto-9609/android_kernel_motorola_exynos9610.git
7 years agoath5k: fix memory leak on buf on failed eeprom read
Colin Ian King [Wed, 3 May 2017 14:26:00 +0000 (15:26 +0100)]
ath5k: fix memory leak on buf on failed eeprom read

The AR5K_EEPROM_READ macro returns with -EIO if a read error
occurs causing a memory leak on the allocated buffer buf. Fix
this by explicitly calling ath5k_hw_nvram_read and exiting on
the via the freebuf label that performs the necessary free'ing
of buf when a read error occurs.

Detected by CoverityScan, CID#1248782 ("Resource Leak")

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath9k: fix spelling in ath9k_tx99_init()
Ammly Fredrick [Thu, 27 Apr 2017 16:31:37 +0000 (19:31 +0300)]
ath9k: fix spelling in ath9k_tx99_init()

It's spelled hardware, not harware.

Signed-off-by: Ammly Fredrick <ammlyf@gmail.com>
[kvalo@qca.qualcomm.com: improve commit log]
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath6kl: assure headroom of skbuff is writable in .start_xmit()
Arend Van Spriel [Tue, 25 Apr 2017 09:10:08 +0000 (10:10 +0100)]
ath6kl: assure headroom of skbuff is writable in .start_xmit()

An issue was found brcmfmac driver in which a skbuff in .start_xmit()
callback was actually cloned. So instead of checking for sufficient
headroom it should also be writable. Hence use skb_cow_head() to
check and expand the headroom appropriately.

Signed-off-by: Arend van Spriel <arend.vanspriel@broadcom.com>
Tested-by: Steve deRosier <derosier@gmail.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath10k: append the wmi_op_version to testmode get_version cmd
Ryan Hsu [Tue, 25 Apr 2017 21:19:16 +0000 (14:19 -0700)]
ath10k: append the wmi_op_version to testmode get_version cmd

QCA9xxx and QCA61x4/QCA93xx are using different wmi operation, in order
for userspace to differentiate it, appends the wmi_op_version information
alone with the get_version command.

Signed-off-by: Ryan Hsu <ryanhsu@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath10k: fix out of bounds access to local buffer
Michael Mera [Mon, 24 Apr 2017 07:11:57 +0000 (16:11 +0900)]
ath10k: fix out of bounds access to local buffer

During write to debugfs file simulate_fw_crash, fixed-size local buffer
'buf' is accessed and modified at index 'count-1', where 'count' is the
size of the write (so potentially out of bounds).
This patch fixes this problem.

Signed-off-by: Michael Mera <dev@michaelmera.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath10k: add initial SDIO support
Erik Stromdahl [Wed, 26 Apr 2017 09:18:00 +0000 (12:18 +0300)]
ath10k: add initial SDIO support

Chipsets like QCA6584 have support for SDIO so add initial SDIO bus support to
ath10k. With this patch we have the low level HTC protocol working and it's
possible to boot the firmware, but it's still not possible to connect or
anything like. More changes are needed for full functionality. For that reason
we print during initialisation:

WARNING: ath10k SDIO support is incomplete, don't expect anything to work!

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
[kvalo@qca.qualcomm.com: refactoring, cleanup, commit log]
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath10k: different fw file name for sdio
Erik Stromdahl [Wed, 26 Apr 2017 09:17:59 +0000 (12:17 +0300)]
ath10k: different fw file name for sdio

Since both SDIO based chipsets will use different
firmware from the PCIe and AHB chipsets, the fw file name
must be different depending on bus type.

The new firmware names are:

For PCIe and AHB:
firmware-<api_version>.bin (same as before)

For SDIO:
firmware-sdio-<api_version>.bin

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath10k: htc: ready_ext msg support
Erik Stromdahl [Wed, 26 Apr 2017 09:17:58 +0000 (12:17 +0300)]
ath10k: htc: ready_ext msg support

Added support for extended ready message.
The extended ready message contains the maximum bundle
count supported by SDIO chipsets.

It is transmitted by SDIO chipset only and replaces the
"standard" ready message in this case.

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath10k: sdio get target info
Erik Stromdahl [Wed, 26 Apr 2017 09:17:57 +0000 (12:17 +0300)]
ath10k: sdio get target info

Special BMI get target info function for SDIO.

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath10k: add sdio extra initializations
Erik Stromdahl [Wed, 26 Apr 2017 09:17:56 +0000 (12:17 +0300)]
ath10k: add sdio extra initializations

Extra initializations needed by all sdio boards.
Derived from qcacld.

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath10k: various sdio related definitions
Erik Stromdahl [Wed, 26 Apr 2017 09:17:55 +0000 (12:17 +0300)]
ath10k: various sdio related definitions

Debug masks for SDIO HIF layer.
Address definitions for SDIO/mbox based chipsets.
Augmented struct host_interest with more members.

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath10k: htc: refactorization
Erik Stromdahl [Wed, 26 Apr 2017 09:17:54 +0000 (12:17 +0300)]
ath10k: htc: refactorization

Code refactorization:

Moved the code for ep 0 in ath10k_htc_rx_completion_handler
to ath10k_htc_control_rx_complete.

This eases the implementation of SDIO/mbox significantly since
the ep_rx_complete cb is invoked directly from the SDIO/mbox
hif layer.

Since the ath10k_htc_control_rx_complete already is present
(only containing a warning message) there is no reason for not
using it (instead of having a special case for ep 0 in
ath10k_htc_rx_completion_handler).

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath10k: htc: move htc ctrl ep connect to htc_init
Erik Stromdahl [Wed, 26 Apr 2017 09:17:53 +0000 (12:17 +0300)]
ath10k: htc: move htc ctrl ep connect to htc_init

This patch moves the HTC ctrl service connect from
htc_wait_target to htc_init.

This is done in order to make sure the htc ctrl service
is setup properly before hif_start is called.

The reason for this is that we want the HTC ctrl service
callback to be initialized before the target sends the
HTC ready message.

The ready message will always be transmitted on endpoint 0
(which is always assigned to the HTC control service) so it
makes more sense if HTC control has been connected before the
ready message is received.

Since the service to pipe mapping is done as a part of
the service connect, the get_default_pipe call is redundant
and was removed.

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath10k: htc: rx trailer lookahead support
Erik Stromdahl [Wed, 26 Apr 2017 09:17:52 +0000 (12:17 +0300)]
ath10k: htc: rx trailer lookahead support

The RX trailer parsing is now capable of parsing lookahead reports.
A lookahead contains the first 4 bytes of the next HTC message
(that will be read in the next SDIO read operation).
Lookaheads are used by the SDIO/mbox HIF layer to determine if
the next message is part of a bundle, which endpoint it belongs
to and how long it is.

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agoath10k: htc: made static function public
Erik Stromdahl [Wed, 26 Apr 2017 09:17:51 +0000 (12:17 +0300)]
ath10k: htc: made static function public

Changed ath10k_htc_notify_tx_completion and
ath10k_htc_process_trailer from static to non static.

These functions are needed by SDIO/mbox.

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agobrcmfmac: Make skb header writable before use
James Hughes [Tue, 25 Apr 2017 09:15:06 +0000 (10:15 +0100)]
brcmfmac: Make skb header writable before use

The driver was making changes to the skb_header without
ensuring it was writable (i.e. uncloned).
This patch also removes some boiler plate header size
checking/adjustment code as that is also handled by the
skb_cow_header function used to make header writable.

Signed-off-by: James Hughes <james.hughes@raspberrypi.org>
Acked-by: Arend van Spriel <arend.vanspriel@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
7 years agobrcmfmac: Ensure pointer correctly set if skb data location changes
James Hughes [Mon, 24 Apr 2017 11:40:50 +0000 (12:40 +0100)]
brcmfmac: Ensure pointer correctly set if skb data location changes

The incoming skb header may be resized if header space is
insufficient, which might change the data adddress in the skb.
Ensure that a cached pointer to that data is correctly set by
moving assignment to after any possible changes.

Signed-off-by: James Hughes <james.hughes@raspberrypi.org>
Acked-by: Arend van Spriel <arend.vanspriel@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
7 years agorndis_wlan: add return value validation
Pan Bian [Mon, 24 Apr 2017 00:40:28 +0000 (08:40 +0800)]
rndis_wlan: add return value validation

Function create_singlethread_workqueue() will return a NULL pointer if
there is no enough memory, and its return value should be validated
before using. However, in function rndis_wlan_bind(), its return value
is not checked. This may cause NULL dereference bugs. This patch fixes
it.

Signed-off-by: Pan Bian <bianpan2016@163.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
7 years agolibertas: check return value of alloc_workqueue
Pan Bian [Sun, 23 Apr 2017 13:19:38 +0000 (21:19 +0800)]
libertas: check return value of alloc_workqueue

Function alloc_workqueue() will return a NULL pointer if there is no
enough memory, and its return value should be validated before using.
However, in function if_spi_probe(), its return value is not checked.
This may result in a NULL dereference bug. This patch fixes the bug.

Signed-off-by: Pan Bian <bianpan2016@163.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
7 years agomt7601u: check return value of alloc_skb
Pan Bian [Sun, 23 Apr 2017 07:00:23 +0000 (15:00 +0800)]
mt7601u: check return value of alloc_skb

Function alloc_skb() will return a NULL pointer if there is no enough
memory. However, in function mt7601u_mcu_msg_alloc(), its return value
is not validated before it is used. This patch fixes it.

Signed-off-by: Pan Bian <bianpan2016@163.com>
Acked-by: Jakub Kicinski <kubakici@wp.pl>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
7 years agoorinoco_usb: Fix buffer on stack
Maksim Salau [Sat, 22 Apr 2017 17:03:06 +0000 (20:03 +0300)]
orinoco_usb: Fix buffer on stack

Allocate buffer on HEAP instead of STACK for a local variable
that is to be sent using usb_control_msg().

Signed-off-by: Maksim Salau <maksim.salau@gmail.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
7 years agoorinoco: fix spelling mistake: "Registerred" -> "Registered"
Colin Ian King [Sat, 22 Apr 2017 13:21:49 +0000 (14:21 +0100)]
orinoco: fix spelling mistake: "Registerred" -> "Registered"

trivial fix to spelling mistake in dbg_dbg message

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
7 years agortlwifi: rtl8821ae: setup 8812ae RFE according to device type
Larry Finger [Mon, 17 Apr 2017 00:32:07 +0000 (19:32 -0500)]
rtlwifi: rtl8821ae: setup 8812ae RFE according to device type

Current channel switch implementation sets 8812ae RFE reg value assuming
that device always has type 2.

Extend possible RFE types set and write corresponding reg values.

Source for new code is
http://dlcdnet.asus.com/pub/ASUS/wireless/PCE-AC51/DR_PCE_AC51_20232801152016.zip

Signed-off-by: Maxim Samoylov <max7255@gmail.com>
Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
Cc: Stable <stable@vger.kernel.org>
Cc: Yan-Hsuan Chuang <yhchuang@realtek.com>
Cc: Pkshih <pkshih@realtek.com>
Cc: Birming Chiu <birming@realtek.com>
Cc: Shaofu <shaofu@realtek.com>
Cc: Steven Ting <steventing@realtek.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
7 years agonet: move xdp_prog field in RX cache lines
Eric Dumazet [Tue, 25 Apr 2017 18:36:52 +0000 (11:36 -0700)]
net: move xdp_prog field in RX cache lines

(struct net_device, xdp_prog) field should be moved in RX cache lines,
reducing latencies when a single packet is received on idle host,
since netif_elide_gro() needs it.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodrivers: net: xgene-v2: Fix error return code in xge_mdio_config()
Wei Yongjun [Tue, 25 Apr 2017 11:36:50 +0000 (11:36 +0000)]
drivers: net: xgene-v2: Fix error return code in xge_mdio_config()

Fix to return error code -ENODEV from the no PHY found error
handling case instead of 0, as done elsewhere in this function.

Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: Generic XDP
David S. Miller [Tue, 18 Apr 2017 19:36:58 +0000 (15:36 -0400)]
net: Generic XDP

This provides a generic SKB based non-optimized XDP path which is used
if either the driver lacks a specific XDP implementation, or the user
requests it via a new IFLA_XDP_FLAGS value named XDP_FLAGS_SKB_MODE.

It is arguable that perhaps I should have required something like
this as part of the initial XDP feature merge.

I believe this is critical for two reasons:

1) Accessibility.  More people can play with XDP with less
   dependencies.  Yes I know we have XDP support in virtio_net, but
   that just creates another depedency for learning how to use this
   facility.

   I wrote this to make life easier for the XDP newbies.

2) As a model for what the expected semantics are.  If there is a pure
   generic core implementation, it serves as a semantic example for
   driver folks adding XDP support.

One thing I have not tried to address here is the issue of
XDP_PACKET_HEADROOM, thanks to Daniel for spotting that.  It seems
incredibly expensive to do a skb_cow(skb, XDP_PACKET_HEADROOM) or
whatever even if the XDP program doesn't try to push headers at all.
I think we really need the verifier to somehow propagate whether
certain XDP helpers are used or not.

v5:
 - Handle both negative and positive offset after running prog
 - Fix mac length in XDP_TX case (Alexei)
 - Use rcu_dereference_protected() in free_netdev (kbuild test robot)

v4:
 - Fix MAC header adjustmnet before calling prog (David Ahern)
 - Disable LRO when generic XDP is installed (Michael Chan)
 - Bypass qdisc et al. on XDP_TX and record the event (Alexei)
 - Do not perform generic XDP on reinjected packets (DaveM)

v3:
 - Make sure XDP program sees packet at MAC header, push back MAC
   header if we do XDP_TX.  (Alexei)
 - Elide GRO when generic XDP is in use.  (Alexei)
 - Add XDP_FLAG_SKB_MODE flag which the user can use to request generic
   XDP even if the driver has an XDP implementation.  (Alexei)
 - Report whether SKB mode is in use in rtnl_xdp_fill() via XDP_FLAGS
   attribute.  (Daniel)

v2:
 - Add some "fall through" comments in switch statements based
   upon feedback from Andrew Lunn
 - Use RCU for generic xdp_prog, thanks to Johannes Berg.

Tested-by: Andy Gospodarek <andy@greyhouse.net>
Tested-by: Jesper Dangaard Brouer <brouer@redhat.com>
Tested-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: fix invalid use of sizeof in qed_alloc_qm_data()
Wei Yongjun [Tue, 25 Apr 2017 07:07:18 +0000 (07:07 +0000)]
qed: fix invalid use of sizeof in qed_alloc_qm_data()

sizeof() when applied to a pointer typed expression gives the
size of the pointer, not that of the pointed data.

Fixes: b5a9ee7cf3be ("qed: Revise QM configuration")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Acked-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: map_get_next_key to return first key on NULL
Teng Qin [Tue, 25 Apr 2017 02:00:37 +0000 (19:00 -0700)]
bpf: map_get_next_key to return first key on NULL

When iterating through a map, we need to find a key that does not exist
in the map so map_get_next_key will give us the first key of the map.
This often requires a lot of guessing in production systems.

This patch makes map_get_next_key return the first key when the key
pointer in the parameter is NULL.

Signed-off-by: Teng Qin <qinteng@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoselftests/net: Fix broken test case in psock_fanout
Mike Maloney [Tue, 25 Apr 2017 01:29:11 +0000 (21:29 -0400)]
selftests/net: Fix broken test case in psock_fanout

The error return falue form sock_fanout_open is -1, not zero.  One test
case was checking for 0 instead of -1.

Tested: Built and tested in clean client.
Signed-off-by: Mike Maloney <maloney@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: netcp_core: remove unused compl queue mapping
Ivan Khoronzhuk [Mon, 24 Apr 2017 20:54:06 +0000 (23:54 +0300)]
net: ethernet: ti: netcp_core: remove unused compl queue mapping

This code is unused and probably was unintentionally left while
moving completion queue mapping in submit function.

Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'qed-vf-tunnel'
David S. Miller [Tue, 25 Apr 2017 15:49:33 +0000 (11:49 -0400)]
Merge branch 'qed-vf-tunnel'

Manish Chopra says:

====================
qed/qede: VF tunnelling support

With this series VFs can run vxlan/geneve/gre tunnels over it.
Please consider applying this series to "net-next"
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed - VF tunnelling support [VXLAN/GENEVE/GRE]
Chopra, Manish [Mon, 24 Apr 2017 17:00:49 +0000 (10:00 -0700)]
qed - VF tunnelling support [VXLAN/GENEVE/GRE]

This patch adds hardware channel APIs support between
VF and PF for tunnelling configuration for the VFs.
According to that configuration VFs can run VXLAN/GENEVE/GRE
tunnels over it with tunnel features offloaded.

Using these APIs VF can also request for UDP ports configuration
to the PF, although PF and it's child VFs share the same port.

Signed-off-by: Manish Chopra <manish.chopra@cavium.com>
Signed-off-by: Yuval Mintz <yuval.mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed/qede: Add UDP ports in bulletin board
Chopra, Manish [Mon, 24 Apr 2017 17:00:48 +0000 (10:00 -0700)]
qed/qede: Add UDP ports in bulletin board

This patch adds support for UDP ports in bulletin board
to notify UDP ports change to the VFs

Signed-off-by: Manish Chopra <manish.chopra@cavium.com>
Signed-off-by: Yuval Mintz <yuval.mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqede: Configure UDP ports in local context.
Chopra, Manish [Mon, 24 Apr 2017 17:00:47 +0000 (10:00 -0700)]
qede: Configure UDP ports in local context.

This patch configures UDP ports locally instead of
configuring them in deferred context which would be
helpful in synchronizing UDP ports configuration for VFs
which will be enabled in further patches.

Signed-off-by: Manish Chopra <manish.chopra@cavium.com>
Signed-off-by: Yuval Mintz <yuval.mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqede: Disable tunnel offloads for non offloaded UDP ports
Chopra, Manish [Mon, 24 Apr 2017 17:00:46 +0000 (10:00 -0700)]
qede: Disable tunnel offloads for non offloaded UDP ports

This patch disables tunnel offloads via ndo_features_check()
if given UDP port is not offloaded to hardware. This in turn
allows to run multiple tunnel interfaces using different UDP ports.

Signed-off-by: Manish Chopra <manish.chopra@cavium.com>
Signed-off-by: Yuval Mintz <yuval.mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed/qede: Enable tunnel offloads based on hw configuration
Chopra, Manish [Mon, 24 Apr 2017 17:00:45 +0000 (10:00 -0700)]
qed/qede: Enable tunnel offloads based on hw configuration

This patch enables tunnel feature offloads based on hw configuration
at initialization time instead of enabling them always.

Signed-off-by: Manish Chopra <manish.chopra@cavium.com>
Signed-off-by: Yuval Mintz <yuval.mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: refactor tunnelling - API/Structs
Chopra, Manish [Mon, 24 Apr 2017 17:00:44 +0000 (10:00 -0700)]
qed: refactor tunnelling - API/Structs

This patch changes the tunnel APIs to use per tunnel
info instead of using bitmasks for all tunnels and also
uses single struct to hold the data to prepare multiple
variant of tunnel configuration ramrods to be sent to the hardware.

Signed-off-by: Manish Chopra <manish.chopra@cavium.com>
Signed-off-by: Yuval Mintz <yuval.mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'l2tpeth-info'
David S. Miller [Tue, 25 Apr 2017 15:41:57 +0000 (11:41 -0400)]
Merge branch 'l2tpeth-info'

Guillaume Nault says:

====================
l2tp: add informations about l2tpeth interfaces in /sys

Patch #1 lets userspace retrieve the naming scheme of an l2tpeth
interface, using /sys/class/net/<iface>/name_assign_type.

Patch #2 adds the DEVTYPE field in /sys/class/net/<iface>/uevent so
that userspace can reliably know if a device is an l2tpeth interface.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agol2tp: define "l2tpeth" device type
Guillaume Nault [Mon, 24 Apr 2017 12:16:07 +0000 (14:16 +0200)]
l2tp: define "l2tpeth" device type

Export type of l2tpeth interfaces to userspace
(/sys/class/net/<iface>/uevent).

Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Acked-by: James Chapman <jchapman@katalix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agol2tp: set name_assign_type for devices created by l2tp_eth.c
Guillaume Nault [Mon, 24 Apr 2017 12:16:06 +0000 (14:16 +0200)]
l2tp: set name_assign_type for devices created by l2tp_eth.c

Export naming scheme used when creating l2tpeth interfaces
(/sys/class/net/<iface>/name_assign_type). This let userspace know if
the device's name has been generated automatically or defined manually.

Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Acked-by: James Chapman <jchapman@katalix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet sched actions: Complete the JUMPX opcode
Jamal Hadi Salim [Sun, 23 Apr 2017 17:17:28 +0000 (13:17 -0400)]
net sched actions: Complete the JUMPX opcode

per discussion at netconf/netdev:
When we have an action that is capable of branching (example a policer),
we can achieve a continuation of the action graph by programming a
"continue" where we find an exact replica of the same filter rule with a lower
priority and the remainder of the action graph. When you have 100s of thousands
of filters which require such a feature it gets very inefficient to do two
lookups.

This patch completes a leftover feature of action codes. Its time has come.

Example below where a user labels packets with a different skbmark on ingress
of a port depending on whether they have/not exceeded the configured rate.
This mark is then used to make further decisions on some egress port.

 #rate control, very low so we can easily see the effect
sudo $TC actions add action police rate 1kbit burst 90k \
conform-exceed pipe/jump 2 index 10
 # skbedit index 11 will be used if the user conforms
sudo $TC actions add action skbedit mark 11 ok index 11
 # skbedit index 12 will be used if the user does not conform
sudo $TC actions add action skbedit mark 12 ok index 12

 #lets bind the user ..
sudo $TC filter add dev $ETH parent ffff: protocol ip prio 8 u32 \
match ip dst 127.0.0.8/32 flowid 1:10 \
action police index 10 \
action skbedit index 11 \
action skbedit index 12

 #run a ping -f and see what happens..
 #
jhs@foobar:~$ sudo $TC -s filter ls dev $ETH parent ffff: protocol ip
filter pref 8 u32
filter pref 8 u32 fh 800: ht divisor 1
filter pref 8 u32 fh 800::800 order 2048 key ht 800 bkt 0 flowid 1:10  (rule hit 2800 success 1005)
  match 7f000008/ffffffff at 16 (success 1005 )
action order 1:  police 0xa rate 1Kbit burst 23440b mtu 2Kb action pipe/jump 2 overhead 0b
ref 2 bind 1 installed 207 sec used 122 sec
Action statistics:
Sent 84420 bytes 1005 pkt (dropped 0, overlimits 721 requeues 0)
backlog 0b 0p requeues 0

action order 2:  skbedit mark 11 pass
 index 11 ref 2 bind 1 installed 204 sec used 122 sec
  Action statistics:
Sent 60564 bytes 721 pkt (dropped 0, overlimits 0 requeues 0)
backlog 0b 0p requeues 0

action order 3:  skbedit mark 12 pass
 index 12 ref 2 bind 1 installed 201 sec used 122 sec
  Action statistics:
Sent 23856 bytes 284 pkt (dropped 0, overlimits 0 requeues 0)
backlog 0b 0p requeues 0

Not bad, about 28% non-conforming packets..

Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge tag 'linux-can-next-for-4.12-20170425' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Tue, 25 Apr 2017 14:52:47 +0000 (10:52 -0400)]
Merge tag 'linux-can-next-for-4.12-20170425' of git://git./linux/kernel/git/mkl/linux-can-next

Marc Kleine-Budde says:

====================
pull-request: can-next 2017-04-25

this is a pull request of 21 patches for net-next/master.

There are 4 patches by Stephane Grosjean for the PEAK PCAN-PCIe FD
CAN-FD boards. The next 7 patches are by Mario Huettel, which add
support for M_CAN IP version >= v3.1.x to the m_can driver. A patch by
Remigiusz Kołłątaj adds support for the Microchip CAN BUS Analyzer. 8
patches by Oliver Hartkopp complete the initial CAN network namespace
support. Wei Yongjun's patch for the ti_hecc driver fixes the return
value check in the probe function.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoipvlan: use pernet operations and restrict l3s hooks to master netns
Florian Westphal [Thu, 20 Apr 2017 16:08:15 +0000 (18:08 +0200)]
ipvlan: use pernet operations and restrict l3s hooks to master netns

commit 4fbae7d83c98c30efc ("ipvlan: Introduce l3s mode") added
registration of netfilter hooks via nf_register_hooks().

This API provides the illusion of 'global' netfilter hooks by placing the
hooks in all current and future network namespaces.

In case of ipvlan the hook appears to be only needed in the namespace
that contains the ipvlan master device (i.e., usually init_net), so
placing them in all namespaces is not needed.

This switches ipvlan driver to pernet operations, and then only registers
hooks in namespaces where a ipvlan master device is set to l3s mode.

Extra care has to be taken when the master device is moved to another
namespace, as we might have to 'move' the netfilter hooks too.

This is done by storing the namespace the ipvlan port was created in.
On REGISTER event, do (un)register operations in the old/new namespaces.

This will also allow removal of the nf_register_hooks() in a future patch.

Cc: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agocan: ti_hecc: fix return value check in ti_hecc_probe()
Wei Yongjun [Tue, 25 Apr 2017 06:44:05 +0000 (06:44 +0000)]
can: ti_hecc: fix return value check in ti_hecc_probe()

In case of error, the function devm_ioremap_resource() returns ERR_PTR()
and never returns NULL. The NULL test in the return value check should
be replaced with IS_ERR().

Fixes: dabf54dd1c63 ("can: ti_hecc: Convert TI HECC driver to DT only driver")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: enable module auto loading for virtual CAN interfaces
Oliver Hartkopp [Tue, 25 Apr 2017 06:19:45 +0000 (08:19 +0200)]
can: enable module auto loading for virtual CAN interfaces

Autoload the vcan module when a vcan instance is to be created by
'ip link add type vcan'

Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: add Virtual CAN Tunnel driver (vxcan)
Oliver Hartkopp [Tue, 25 Apr 2017 06:19:44 +0000 (08:19 +0200)]
can: add Virtual CAN Tunnel driver (vxcan)

Similar to the virtual ethernet driver veth, vxcan implements a
local CAN traffic tunnel between two virtual CAN network devices.
See Kconfig entry for details.

Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: network namespace support for CAN gateway
Oliver Hartkopp [Tue, 25 Apr 2017 06:19:43 +0000 (08:19 +0200)]
can: network namespace support for CAN gateway

The CAN gateway was not implemented as per-net in the initial network
namespace support by Mario Kicherer (8e8cda6d737d).
This patch enables the CAN gateway to be used in different namespaces.

Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: network namespace support for CAN_BCM protocol
Oliver Hartkopp [Tue, 25 Apr 2017 06:19:42 +0000 (08:19 +0200)]
can: network namespace support for CAN_BCM protocol

The CAN_BCM protocol and its procfs entries were not implemented as per-net
in the initial network namespace support by Mario Kicherer (8e8cda6d737d).
This patch adds the missing per-net functionality for the CAN BCM.

Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: complete initial namespace support
Oliver Hartkopp [Tue, 25 Apr 2017 06:19:41 +0000 (08:19 +0200)]
can: complete initial namespace support

The statistics and its proc output was not implemented as per-net in the
initial network namespace support by Mario Kicherer (8e8cda6d737d).
This patch adds the missing per-net statistics for the CAN subsystem.

Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: remove obsolete definitions
Oliver Hartkopp [Tue, 25 Apr 2017 06:19:40 +0000 (08:19 +0200)]
can: remove obsolete definitions

can_rx_alldev_list is a per-net data structure now. Remove it's definition
here and can_rx_dev_list too.

Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: remove obsolete pernet_operations definitions
Oliver Hartkopp [Tue, 25 Apr 2017 06:19:39 +0000 (08:19 +0200)]
can: remove obsolete pernet_operations definitions

The namespace support for the CAN subsystem does not need any additional
memory. So when ".size = 0" there's no extra memory allocated by the system.
And therefore ".id" is obsolete too.

Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: fix memory leak in initial namespace support
Oliver Hartkopp [Tue, 25 Apr 2017 06:19:38 +0000 (08:19 +0200)]
can: fix memory leak in initial namespace support

The can_rx_alldev_list is a per-net data structure now and allocated in
can_pernet_init(). Make sure the memory is free'd in can_pernet_exit() too.

Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: mcba_usb: Add support for Microchip CAN BUS Analyzer
Remigiusz Kołłątaj [Fri, 14 Apr 2017 18:32:28 +0000 (20:32 +0200)]
can: mcba_usb: Add support for Microchip CAN BUS Analyzer

SocketCAN driver for Microchip CAN BUS Analyzer
(http://www.microchip.com/development-tools/)

Changes in v4:
- possible memory leak fixed in mcba_usb_write_bulk_callback
- LED support added
- failure handling in mcba_usb_probe improved
- C99 initializers for structs on stack

Changes in v3:
- improved/simplified CAN ID conversion
- functions for transmission of skb and cmd separated
- fixed/improved netif_stop_queue handling
- style/cosmetic corrections

Changes in v2:
- Termination handling reimplemented to fit new netlink API
(IFLA_CAN_TERMINATION)
- Bitrate handling reimplemented to fit new netlink API
(IFLA_CAN_BITRATE)
- CAN ID conversion refactored (changed from macro to inline functions)
- CAN DLC handling using get_can_dlc()
- Endianness handling for can_speed introduced
- Debugging removed
- Redundant error prints removed
- Style/cosmetic corrections (i.e. macro names, redefs, inits etc.)

Signed-off-by: Remigiusz Kołłątaj <remigiusz.kollataj@mobica.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: m_can: Enable TX FIFO Handling for M_CAN IP version >= v3.1.x
Mario Huettel [Sat, 8 Apr 2017 12:10:15 +0000 (14:10 +0200)]
can: m_can: Enable TX FIFO Handling for M_CAN IP version >= v3.1.x

* Added defines for TX Event FIFO Element
* Adapted ndo_start_xmit function.
  For versions >= v3.1.x it uses the TX FIFO to optimize the data
  throughput. It stores the echo skb at the same index as in the
  M_CAN's TX FIFO. The frame's message marker is set to this index.
  This message marker is received in the TX Event FIFO after
  the message was successfully transmitted. It is used to echo the
  correct echo skb back to the network stack.
* Added m_can_echo_tx_event function. It reads all received
  message markers in the TX Event FIFO and loops back the
  corresponding echo skbs.
* ISR checks for new TX Event Entry interrupt for version >= 3.1.x.

Signed-off-by: Mario Huettel <mario.huettel@gmx.net>
Reviewed-by: Oliver Hartkopp <socketcan@hartkopp.net>
Tested-by: Quentin Schulz <quentin.schulz@free-electrons.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: m_can: Configuration for TX and TX event FIFOs
Mario Huettel [Sat, 8 Apr 2017 12:10:14 +0000 (14:10 +0200)]
can: m_can: Configuration for TX and TX event FIFOs

* TX/TX Event FIFO sizes are configured for version >= v3.1.x

Signed-off-by: Mario Huettel <mario.huettel@gmx.net>
Tested-by: Quentin Schulz <quentin.schulz@free-electrons.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: m_can: Enable M_CAN version dependent initialization
Mario Huettel [Sat, 8 Apr 2017 12:10:13 +0000 (14:10 +0200)]
can: m_can: Enable M_CAN version dependent initialization

This patch adapts the initialization of the M_CAN. So it can be used
with all versions >= 3.0.x.

Changes:
* Added version element to m_can_priv structure to hold M_CAN version.
* Renamed bittiming structs for version 3.0.x
* Added new bittiming structs for version >= 3.1.x
* Function alloc_m_can_dev takes 2 new arguments. The TX FIFO size and the
  base address of the module.
* Chip configuration for CAN_CTRLMODE_LOOPBACK is changed: Enabled
  CCCR_MON bit. In combination with TEST_LBCK it activates the internal
  loopback mode. Leaving CCCR_MON '0' results in external loopback mode.
* Clocks are temporarily enabled by platform_propbe function in order to
  allow read access to the Core Release register and the Control Register.
  Registers are used to detect M_CAN version and optional Non-ISO Feature.

Initialization of M_CAN for version >= 3.1.x:
* TX FIFO of M_CAN is used to transmit frames. The driver does not need to
  stop the tx queue after each frame sent.
* Initialization of TX Event FIFO is added.
* NON-ISO is fixed for all M_CAN versions < 3.2.x. Version 3.2.x _can_ have
  the NISO (Non-ISO) bit which can switch the mode of the M_CAN to Non-ISO
  mode. This bit does not have to be writeable. Therefore it is checked.
  If it is writable Non-ISO support is added to the controllers supported
  CAN modes.

New Functions:
* Function to check the Core Release version. The read value determines the
  behaviour of the driver.
* Function to check if the NISO bit for version >= 3.2.x is implemented.

Signed-off-by: Mario Huettel <mario.huettel@gmx.net>
Reviewed-by: Oliver Hartkopp <socketcan@hartkopp.net>
Tested-by: Quentin Schulz <quentin.schulz@free-electrons.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: m_can: Updated register defines to newest version
Mario Huettel [Sat, 8 Apr 2017 12:10:12 +0000 (14:10 +0200)]
can: m_can: Updated register defines to newest version

* Updated register defines to newest M_CAN version (v3.2.1).
* Changed defines in the whole code.

Signed-off-by: Mario Huettel <mario.huettel@gmx.net>
Reviewed-by: Oliver Hartkopp <socketcan@hartkopp.net>
Tested-by: Quentin Schulz <quentin.schulz@free-electrons.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: m_can: Removed virtual address from print
Mario Huettel [Sat, 8 Apr 2017 12:10:11 +0000 (14:10 +0200)]
can: m_can: Removed virtual address from print

The virtual address of the device was printed. I removed it because it
leaks internal information.

Signed-off-by: Mario Huettel <mario.huettel@gmx.net>
Tested-by: Quentin Schulz <quentin.schulz@free-electrons.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: m_can: Removed initialization of FIFO water marks
Mario Huettel [Sat, 8 Apr 2017 12:10:10 +0000 (14:10 +0200)]
can: m_can: Removed initialization of FIFO water marks

FIFO water marks disabled because the driver doesn't handle water mark
events.

Signed-off-by: Mario Huettel <mario.huettel@gmx.net>
Reviewed-by: Oliver Hartkopp <socketcan@hartkopp.net>
Tested-by: Quentin Schulz <quentin.schulz@free-electrons.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: m_can: Disabled Interrupt Line 1
Mario Huettel [Sat, 8 Apr 2017 12:10:09 +0000 (14:10 +0200)]
can: m_can: Disabled Interrupt Line 1

* Disabled interrupt line 1. The driver didn't use it.

Signed-off-by: Mario Huettel <mario.huettel@gmx.net>
Tested-by: Quentin Schulz <quentin.schulz@free-electrons.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: peak: add support for PEAK PCAN-PCIe FD CAN-FD boards
Stephane Grosjean [Thu, 19 Jan 2017 15:31:07 +0000 (16:31 +0100)]
can: peak: add support for PEAK PCAN-PCIe FD CAN-FD boards

This patch adds the support of the PCAN-PCI Express FD boards made
by PEAK-System, for computers using the PCI Express slot.

The PCAN-PCI Express FD has one or two CAN FD channels, depending
on the model. A galvanic isolation of the CAN ports protects
the electronics of the card and the respective computer against
disturbances of up to 500 Volts. The PCAN-PCI Express FD can be operated
with ambient temperatures in a range of -40 to +85 °C.

Such boards run an extented version of the CAN-FD IP running into USB
CAN-FD interfaces from PEAK-System, so this patch adds several new commands
and their corresponding data types to the PEAK CAN-FD common definitions
header file too.

Signed-off-by: Stephane Grosjean <s.grosjean@peak-system.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: peak: move header file to new can common subdir
Stephane Grosjean [Thu, 19 Jan 2017 15:31:06 +0000 (16:31 +0100)]
can: peak: move header file to new can common subdir

The CAN-FD IP from PEAK-System runs into several kinds of PC CAN-FD
interfaces. Up to now, only the USB CAN-FD adapters were supported by
the Kernel. In order to prepare the adding of some new non-USB CAN-FD
interfaces, this patch moves - and rename - the IP definitions file
from its private (usb) sub-directory into a - newly created - CAN specific
one.

Signed-off-by: Stephane Grosjean <s.grosjean@peak-system.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: peak: fix usage of const qualifier in pointers args
Stephane Grosjean [Thu, 19 Jan 2017 15:31:05 +0000 (16:31 +0100)]
can: peak: fix usage of const qualifier in pointers args

Fixes the usage of the const qualifier in the memory pointer arguments
of the declared inline functions. By changing the line containing "const",
this patch also changes the name of the arg into a more usual one.

Signed-off-by: Stephane Grosjean <s.grosjean@peak-system.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agocan: peak: fix usage of usb specific data type
Stephane Grosjean [Thu, 19 Jan 2017 15:31:04 +0000 (16:31 +0100)]
can: peak: fix usage of usb specific data type

This patch fixes the wrong usage of a specific USB data type into a common
header file. This common header file is intended to define the common data
types and values that define access to the PEAK-System CAN-FD IP, whatever
the PC interface is.

Signed-off-by: Stephane Grosjean <s.grosjean@peak-system.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
7 years agoMerge branch 'virtio-net-tx-napi'
David S. Miller [Tue, 25 Apr 2017 03:55:20 +0000 (23:55 -0400)]
Merge branch 'virtio-net-tx-napi'

Willem de Bruijn says:

====================
virtio-net tx napi

Add napi for virtio-net transmit completion processing.

Changes:
  v2 -> v3:
    - convert __netif_tx_trylock to __netif_tx_lock on tx napi poll
          ensure that the handler always cleans, to avoid deadlock
    - unconditionally clean in start_xmit
          avoid adding an unnecessary "if (use_napi)" branch
    - remove virtqueue_disable_cb in patch 5/5
          a noop in the common event_idx based loop
    - document affinity_hint_set constraint

  v1 -> v2:
    - disable by default
    - disable unless affinity_hint_set
          because cache misses add up to a third higher cycle cost,
  e.g., in TCP_RR tests. This is not limited to the patch
  that enables tx completion cleaning in rx napi.
    - use trylock to avoid contention between tx and rx napi
    - keep interrupts masked during xmit_more (new patch 5/5)
          this improves cycles especially for multi UDP_STREAM, which
  does not benefit from cleaning tx completions on rx napi.
    - move free_old_xmit_skbs (new patch 3/5)
          to avoid forward declaration

    not changed:
    - deduplicate virnet_poll_tx and virtnet_poll_txclean
          they look similar, but have differ too much to make it
  worthwhile.
    - delay netif_wake_subqueue for more than 2 + MAX_SKB_FRAGS
          evaluated, but made no difference
    - patch 1/5

  RFC -> v1:
    - dropped vhost interrupt moderation patch:
          not needed and likely expensive at light load
    - remove tx napi weight
        - always clean all tx completions
        - use boolean to toggle tx-napi, instead
    - only clean tx in rx if tx-napi is enabled
        - then clean tx before rx
    - fix: add missing braces in virtnet_freeze_down
    - testing: add 4KB TCP_RR + UDP test results

Based on previous patchsets by Jason Wang:

  [RFC V7 PATCH 0/7] enable tx interrupts for virtio-net
  http://lkml.iu.edu/hypermail/linux/kernel/1505.3/00245.html

Before commit b0c39dbdc204 ("virtio_net: don't free buffers in xmit
ring") the virtio-net driver would free transmitted packets on
transmission of new packets in ndo_start_xmit and, to catch the edge
case when no new packet is sent, also in a timer at 10HZ.

A timer can cause long stalls. VIRTIO_F_NOTIFY_ON_EMPTY avoids stalls
due to low free descriptor count. It does not address a stalls due to
low socket SO_SNDBUF. Increasing timer frequency decreases that stall
time, but increases interrupt rate and, thus, cycle count.

Currently, with no timer, packets are freed only at ndo_start_xmit.
Latency of consume_skb is now unbounded. To avoid a deadlock if a sock
reaches SO_SNDBUF, packets are orphaned on tx. This breaks TCP small
queues.

Reenable TCP small queues by removing the orphan. Instead of using a
timer, convert the driver to regular tx napi. This does not have the
unresolved stall issue and does not have any frequency to tune.

By keeping interrupts enabled by default, napi increases tx
interrupt rate. VIRTIO_F_EVENT_IDX avoids sending an interrupt if
one is already unacknowledged, so makes this more feasible today.
Combine that with an optimization that brings interrupt rate
back in line with the existing version for most workloads:

Tx completion cleaning on rx interrupts elides most explicit tx
interrupts by relying on the fact that many rx interrupts fire.

Tested by running {1, 10, 100} {TCP, UDP} STREAM, RR, 4K_RR benchmarks
from a guest to a server on the host, on an x86_64 Haswell. The guest
runs 4 vCPUs pinned to 4 cores. vhost and the test server are
pinned to a core each.

All results are the median of 5 runs, with variance well < 10%.
Used neper (github.com/google/neper) as test process.

Napi increases single stream throughput, but increases cycle cost.
The optimizations bring this down. The previous patchset saw a
regression with UDP_STREAM, which does not benefit from cleaning tx
interrupts in rx napi. This regression is now gone for 10x, 100x.
Remaining difference is higher 1x TCP_STREAM, lower 1x UDP_STREAM.

The latest results are with process, rx napi and tx napi affine to
the same core. All numbers are lower than the previous patchset.

             upstream     napi
TCP_STREAM:
1x:
  Mbps          27816    39805
  Gcycles         274      285

10x:
  Mbps          42947    42531
  Gcycles         300      296

100x:
  Mbps          31830    28042
  Gcycles         279      269

TCP_RR Latency (us):
1x:
  p50              21       21
  p99              27       27
  Gcycles         180      167

10x:
  p50              40       39
  p99              52       52
  Gcycles         214      211

100x:
  p50             281      241
  p99             411      337
  Gcycles         218      226

TCP_RR 4K:
1x:
  p50              28       29
  p99              34       36
  Gcycles         177      167

10x:
  p50              70       71
  p99              85      134
  Gcycles         213      214

100x:
  p50             442      611
  p99             802      785
  Gcycles         237      216

UDP_STREAM:
1x:
  Mbps          29468    26800
  Gcycles         284      293

10x:
  Mbps          29891    29978
  Gcycles         285      312

100x:
  Mbps          30269    30304
  Gcycles         318      316

UDP_RR:
1x:
  p50              19       19
  p99              23       23
  Gcycles         180      173

10x:
  p50              35       40
  p99              54       64
  Gcycles         245      237

100x:
  p50             234      286
  p99             484      473
  Gcycles         224      214

Note that GSO is enabled, so 4K RR still translates to one packet
per request.

Lower throughput at 100x vs 10x can be (at least in part)
explained by looking at bytes per packet sent (nstat). It likely
also explains the lower throughput of 1x for some variants.

upstream:

 N=1   bytes/pkt=16581
 N=10  bytes/pkt=61513
 N=100 bytes/pkt=51558

at_rx:

 N=1   bytes/pkt=65204
 N=10  bytes/pkt=65148
 N=100 bytes/pkt=56840
====================

Acked-by: Michael S. Tsirkin <mst@redhat.com>
7 years agovirtio-net: keep tx interrupts disabled unless kick
Willem de Bruijn [Mon, 24 Apr 2017 17:49:30 +0000 (13:49 -0400)]
virtio-net: keep tx interrupts disabled unless kick

Tx napi mode increases the rate of transmit interrupts. Suppress some
by masking interrupts while more packets are expected. The interrupts
will be reenabled before the last packet is sent.

This optimization reduces the througput drop with tx napi for
unidirectional flows such as UDP_STREAM that do not benefit from
cleaning tx completions in the the receive napi handler.

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovirtio-net: clean tx descriptors from rx napi
Willem de Bruijn [Mon, 24 Apr 2017 17:49:29 +0000 (13:49 -0400)]
virtio-net: clean tx descriptors from rx napi

Amortize the cost of virtual interrupts by doing both rx and tx work
on reception of a receive interrupt if tx napi is enabled. With
VIRTIO_F_EVENT_IDX, this suppresses most explicit tx completion
interrupts for bidirectional workloads.

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovirtio-net: move free_old_xmit_skbs
Willem de Bruijn [Mon, 24 Apr 2017 17:49:28 +0000 (13:49 -0400)]
virtio-net: move free_old_xmit_skbs

An upcoming patch will call free_old_xmit_skbs indirectly from
virtnet_poll. Move the function above this to avoid having to
introduce a forward declaration.

This is a pure move: no code changes.

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovirtio-net: transmit napi
Willem de Bruijn [Mon, 24 Apr 2017 17:49:27 +0000 (13:49 -0400)]
virtio-net: transmit napi

Convert virtio-net to a standard napi tx completion path. This enables
better TCP pacing using TCP small queues and increases single stream
throughput.

The virtio-net driver currently cleans tx descriptors on transmission
of new packets in ndo_start_xmit. Latency depends on new traffic, so
is unbounded. To avoid deadlock when a socket reaches its snd limit,
packets are orphaned on tranmission. This breaks socket backpressure,
including TSQ.

Napi increases the number of interrupts generated compared to the
current model, which keeps interrupts disabled as long as the ring
has enough free descriptors. Keep tx napi optional and disabled for
now. Follow-on patches will reduce the interrupt cost.

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovirtio-net: napi helper functions
Willem de Bruijn [Mon, 24 Apr 2017 17:49:26 +0000 (13:49 -0400)]
virtio-net: napi helper functions

Prepare virtio-net for tx napi by converting existing napi code to
use helper functions. This also deduplicates some logic.

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosparc64: Improve 64-bit constant loading in eBPF JIT.
David S. Miller [Tue, 25 Apr 2017 02:42:34 +0000 (19:42 -0700)]
sparc64: Improve 64-bit constant loading in eBPF JIT.

Doing a full 64-bit decomposition is really stupid especially for
simple values like 0 and -1.

But if we are going to optimize this, go all the way and try for all 2
and 3 instruction sequences not requiring a temporary register as
well.

First we do the easy cases where it's a zero or sign extended 32-bit
number (sethi+or, sethi+xor, respectively).

Then we try to find a range of set bits we can load simply then shift
up into place, in various ways.

Then we try negating the constant and see if we can do a simple
sequence using that with a xor at the end.  (f.e. the range of set
bits can't be loaded simply, but for the negated value it can)

The final optimized strategy involves 4 instructions sequences not
needing a temporary register.

Otherwise we sadly fully decompose using a temp..

Example, from ALU64_XOR_K: 0x0000ffffffff0000 ^ 0x0 = 0x0000ffffffff0000:

0000000000000000 <foo>:
   0:   9d e3 bf 50     save  %sp, -176, %sp
   4:   01 00 00 00     nop
   8:   90 10 00 18     mov  %i0, %o0
   c:   13 3f ff ff     sethi  %hi(0xfffffc00), %o1
  10:   92 12 63 ff     or  %o1, 0x3ff, %o1     ! ffffffff <foo+0xffffffff>
  14:   93 2a 70 10     sllx  %o1, 0x10, %o1
  18:   15 3f ff ff     sethi  %hi(0xfffffc00), %o2
  1c:   94 12 a3 ff     or  %o2, 0x3ff, %o2     ! ffffffff <foo+0xffffffff>
  20:   95 2a b0 10     sllx  %o2, 0x10, %o2
  24:   92 1a 60 00     xor  %o1, 0, %o1
  28:   12 e2 40 8a     cxbe  %o1, %o2, 38 <foo+0x38>
  2c:   9a 10 20 02     mov  2, %o5
  30:   10 60 00 03     b,pn   %xcc, 3c <foo+0x3c>
  34:   01 00 00 00     nop
  38:   9a 10 20 01     mov  1, %o5     ! 1 <foo+0x1>
  3c:   81 c7 e0 08     ret
  40:   91 eb 40 00     restore  %o5, %g0, %o0

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosparc64: Support cbcond instructions in eBPF JIT.
David S. Miller [Mon, 24 Apr 2017 22:56:21 +0000 (15:56 -0700)]
sparc64: Support cbcond instructions in eBPF JIT.

cbcond combines a compare with a branch into a single instruction.

The limitations are:

1) Only newer chips support it

2) For immediate compares we are limited to 5-bit signed immediate
   values

3) The branch displacement is limited to 10-bit signed

4) We cannot use it for JSET

Also, cbcond (unlike all other sparc control transfers) lacks a delay
slot.

Currently we don't have a useful instruction we can push into the
delay slot of normal branches.  So using cbcond pretty much always
increases code density, and is therefore a win.

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'bpf-misc-cleanups'
David S. Miller [Mon, 24 Apr 2017 20:20:20 +0000 (16:20 -0400)]
Merge branch 'bpf-misc-cleanups'

Alexander Alemayhu says:

====================
Misc BPF cleanup

while looking into making the Makefile in samples/bpf better handle O= I saw
several warnings when running `make clean && make samples/bpf/`. This series
reduces those warnings.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosamples/bpf: check before defining offsetof
Alexander Alemayhu [Mon, 24 Apr 2017 13:31:08 +0000 (15:31 +0200)]
samples/bpf: check before defining offsetof

Fixes the following warning

samples/bpf/test_lru_dist.c:28:0: warning: "offsetof" redefined
 #define offsetof(TYPE, MEMBER) ((size_t)&((TYPE *)0)->MEMBER)

In file included from ./tools/lib/bpf/bpf.h:25:0,
                 from samples/bpf/libbpf.h:5,
                 from samples/bpf/test_lru_dist.c:24:
/usr/lib/gcc/x86_64-redhat-linux/6.3.1/include/stddef.h:417:0: note: this is the location of the previous definition
 #define offsetof(TYPE, MEMBER) __builtin_offsetof (TYPE, MEMBER)

Signed-off-by: Alexander Alemayhu <alexander@alemayhu.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosamples/bpf: add static to function with no prototype
Alexander Alemayhu [Mon, 24 Apr 2017 13:31:07 +0000 (15:31 +0200)]
samples/bpf: add static to function with no prototype

Fixes the following warning

samples/bpf/cookie_uid_helper_example.c: At top level:
samples/bpf/cookie_uid_helper_example.c:276:6: warning: no previous prototype for ‘finish’ [-Wmissing-prototypes]
 void finish(int ret)
      ^~~~~~
  HOSTLD  samples/bpf/per_socket_stats_example

Signed-off-by: Alexander Alemayhu <alexander@alemayhu.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosamples/bpf: add -Wno-unknown-warning-option to clang
Alexander Alemayhu [Mon, 24 Apr 2017 13:31:06 +0000 (15:31 +0200)]
samples/bpf: add -Wno-unknown-warning-option to clang

I was initially going to remove '-Wno-address-of-packed-member' because I
thought it was not supposed to be there but Daniel suggested using
'-Wno-unknown-warning-option'.

This silences several warnings similiar to the one below

warning: unknown warning option '-Wno-address-of-packed-member' [-Wunknown-warning-option]
1 warning generated.
clang  -nostdinc -isystem /usr/lib/gcc/x86_64-redhat-linux/6.3.1/include -I./arch/x86/include -I./arch/x86/include/generated/uapi -I./arch/x86/include/generated  -I./include
 -I./arch/x86/include/uapi -I./include/uapi -I./include/generated/uapi -include ./include/linux/kconfig.h  \
        -D__KERNEL__ -D__ASM_SYSREG_H -Wno-unused-value -Wno-pointer-sign \
        -Wno-compare-distinct-pointer-types \
        -Wno-gnu-variable-sized-type-not-at-end \
        -Wno-address-of-packed-member -Wno-tautological-compare \
        -O2 -emit-llvm -c samples/bpf/xdp_tx_iptunnel_kern.c -o -| llc -march=bpf -filetype=obj -o samples/bpf/xdp_tx_iptunnel_kern.o

$ clang --version

 clang version 3.9.1 (tags/RELEASE_391/final)
 Target: x86_64-unknown-linux-gnu
 Thread model: posix
 InstalledDir: /usr/bin

Signed-off-by: Alexander Alemayhu <alexander@alemayhu.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: make bpf_xdp_adjust_head support mandatory
Daniel Borkmann [Mon, 24 Apr 2017 20:14:35 +0000 (22:14 +0200)]
bpf: make bpf_xdp_adjust_head support mandatory

Now that also the last in-tree user of the xdp_adjust_head bit has
been removed, we can remove the flag from struct bpf_prog altogether.

This, at the same time, also makes sure that any future driver for
XDP comes with bpf_xdp_adjust_head() support right away.

A rejection based on this flag would also mean that tail calls
couldn't be used with such driver as per c2002f983767 ("bpf: fix
checking xdp_adjust_head on tail calls") fix, thus lets not allow
for it in the first place.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqlcnic: fix unchecked return value
Pan Bian [Sun, 23 Apr 2017 12:04:04 +0000 (20:04 +0800)]
qlcnic: fix unchecked return value

Function pci_find_ext_capability() may return 0, which is an invalid
address. In function qlcnic_sriov_virtid_fn(), its return value is used
without validation. This may result in invalid memory access bugs. This
patch fixes the bug.

Signed-off-by: Pan Bian <bianpan2016@163.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agowan: pc300too: abort path on failure
Pan Bian [Sun, 23 Apr 2017 09:38:35 +0000 (17:38 +0800)]
wan: pc300too: abort path on failure

In function pc300_pci_init_one(), on the ioremap error path, function
pc300_pci_remove_one() is called to free the allocated memory. However,
the path is not terminated, and the freed memory will be used later,
resulting in use-after-free bugs. This path fixes the bug.

Signed-off-by: Pan Bian <bianpan2016@163.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotipc: check return value of nlmsg_new
Pan Bian [Sun, 23 Apr 2017 07:09:19 +0000 (15:09 +0800)]
tipc: check return value of nlmsg_new

Function nlmsg_new() will return a NULL pointer if there is no enough
memory, and its return value should be checked before it is used.
However, in function tipc_nl_node_get_monitor(), the validation of the
return value of function nlmsg_new() is missed. This patch fixes the
bug.

Signed-off-by: Pan Bian <bianpan2016@163.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agolwtunnel: check return value of nla_nest_start
Pan Bian [Sun, 23 Apr 2017 06:28:37 +0000 (14:28 +0800)]
lwtunnel: check return value of nla_nest_start

Function nla_nest_start() may return a NULL pointer on error. However,
in function lwtunnel_fill_encap(), the return value of nla_nest_start()
is not validated before it is used. This patch checks the return value
of nla_nest_start() against NULL.

Signed-off-by: Pan Bian <bianpan2016@163.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'nfp-dma-adjust_head-fixes'
David S. Miller [Mon, 24 Apr 2017 18:35:45 +0000 (14:35 -0400)]
Merge branch 'nfp-dma-adjust_head-fixes'

Jakub Kicinski says:

====================
nfp: DMA flags, adjust head and fixes

This series takes advantage of Alex's DMA_ATTR_SKIP_CPU_SYNC to make
XDP packet modifications "correct" from DMA API point of view.  It
also allows us to parse the metadata before we run XDP at no additional
DMA sync cost.  That way we can get rid of the metadata memcpy, and
remove the last upstream user of bpf_prog->xdp_adjust_head.

David's patch adds a way to read capabilities from the management
firmware.

There are also two net-next fixes.  Patch 4 which fixes what seems to
be a result of a botched rebase on my part.  Patch 5 corrects locking
when state of ethernet ports is being refreshed.

v3: move the sync from alloc func to the actual give to hw func
v2: sync rx buffers before giving them to the card (Alex)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: remove the refresh of all ports optimization
Jakub Kicinski [Sun, 23 Apr 2017 03:17:56 +0000 (20:17 -0700)]
nfp: remove the refresh of all ports optimization

The code refreshing the eth port state was trying to update state
of all ports of the card.  Unfortunately to safely walk the port
list we would have to hold the port lock, which we can't due to
lock ordering constraints against rtnl.

Make the per-port sync refresh and async refresh of all ports
completely separate routines.

Fixes: 172f638c93dd ("nfp: add port state refresh")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: fix free list buffer size reporting
Jakub Kicinski [Sun, 23 Apr 2017 03:17:55 +0000 (20:17 -0700)]
nfp: fix free list buffer size reporting

XDP headroom should not be included in free list buffer size.

Fixes: 6fe0c3b43804 ("nfp: add support for xdp_adjust_head()")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: add NSP routine to get static information
David Brunecz [Sun, 23 Apr 2017 03:17:54 +0000 (20:17 -0700)]
nfp: add NSP routine to get static information

Retrieve identifying information from the NSP.  For now it only
contains versions of firmware subcomponents.

Signed-off-by: David Brunecz <david.brunecz@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: parse metadata prepend before XDP runs
Jakub Kicinski [Sun, 23 Apr 2017 03:17:53 +0000 (20:17 -0700)]
nfp: parse metadata prepend before XDP runs

Calling memcpy to shift metadata out of the way for XDP to run
seems like an overkill.  The most common metadata contents are
8 bytes containing type and flow hash.  Simply parse the metadata
before we run XDP.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: make use of the DMA_ATTR_SKIP_CPU_SYNC attr
Jakub Kicinski [Sun, 23 Apr 2017 03:17:52 +0000 (20:17 -0700)]
nfp: make use of the DMA_ATTR_SKIP_CPU_SYNC attr

DMA unmap may destroy changes CPU made to the buffer.  To make XDP
run correctly on non-x86 platforms we should use the
DMA_ATTR_SKIP_CPU_SYNC attribute.

Thanks to using the attribute we can now push the sync operation to the
common code path from XDP handler.

A little bit of variable name reshuffling is required to bring the
code back to readable state.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'cls_flower-MPLS'
David S. Miller [Mon, 24 Apr 2017 18:30:47 +0000 (14:30 -0400)]
Merge branch 'cls_flower-MPLS'

Benjamin LaHaise says:

====================
flower: add MPLS matching support

This patch series adds support for parsing MPLS flows in the flow dissector
and the flower classifier.  Each of the MPLS TTL, BOS, TC and Label fields
can be used for matching.

v2: incorporate style feedback, move #defines to linux/include/mpls.h
Note: this omits Jiri's request to remove tabs between the type and
field names in struct declarations.  This would be inconsistent with
numerous other struct definitions.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agocls_flower: add support for matching MPLS fields (v2)
Benjamin LaHaise [Sat, 22 Apr 2017 20:52:47 +0000 (16:52 -0400)]
cls_flower: add support for matching MPLS fields (v2)

Add support to the tc flower classifier to match based on fields in MPLS
labels (TTL, Bottom of Stack, TC field, Label).

Signed-off-by: Benjamin LaHaise <benjamin.lahaise@netronome.com>
Signed-off-by: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Simon Horman <simon.horman@netronome.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Jiri Pirko <jiri@mellanox.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Hadar Hen Zion <hadarh@mellanox.com>
Cc: Gao Feng <fgao@ikuai8.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoflow_dissector: add mpls support (v2)
Benjamin LaHaise [Sat, 22 Apr 2017 20:52:46 +0000 (16:52 -0400)]
flow_dissector: add mpls support (v2)

Add support for parsing MPLS flows to the flow dissector in preparation for
adding MPLS match support to cls_flower.

Signed-off-by: Benjamin LaHaise <benjamin.lahaise@netronome.com>
Signed-off-by: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Simon Horman <simon.horman@netronome.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Jiri Pirko <jiri@mellanox.com>
Cc: Eric Dumazet <jhs@mojatatu.com>
Cc: Hadar Hen Zion <hadarh@mellanox.com>
Cc: Gao Feng <fgao@ikuai8.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'tcp-fastopen-middlebox-fixes'
David S. Miller [Mon, 24 Apr 2017 18:27:18 +0000 (14:27 -0400)]
Merge branch 'tcp-fastopen-middlebox-fixes'

Wei Wang says:

====================
net/tcp_fastopen: Fix for various TFO firewall issues

Currently there are still some firewall issues in the middlebox
which make the middlebox drop packets silently for TFO sockets.
This kind of issue is hard to be detected by the end client.

This patch series tries to detect such issues in the kernel and disable
TFO temporarily.
More details about the issues and the fixes are included in the following
patches.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/tcp_fastopen: Remove mss check in tcp_write_timeout()
Wei Wang [Thu, 20 Apr 2017 21:45:48 +0000 (14:45 -0700)]
net/tcp_fastopen: Remove mss check in tcp_write_timeout()

Christoph Paasch from Apple found another firewall issue for TFO:
After successful 3WHS using TFO, server and client starts to exchange
data. Afterwards, a 10s idle time occurs on this connection. After that,
firewall starts to drop every packet on this connection.

The fix for this issue is to extend existing firewall blackhole detection
logic in tcp_write_timeout() by removing the mss check.

Signed-off-by: Wei Wang <weiwan@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/tcp_fastopen: Add snmp counter for blackhole detection
Wei Wang [Thu, 20 Apr 2017 21:45:47 +0000 (14:45 -0700)]
net/tcp_fastopen: Add snmp counter for blackhole detection

This counter records the number of times the firewall blackhole issue is
detected and active TFO is disabled.

Signed-off-by: Wei Wang <weiwan@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/tcp_fastopen: Disable active side TFO in certain scenarios
Wei Wang [Thu, 20 Apr 2017 21:45:46 +0000 (14:45 -0700)]
net/tcp_fastopen: Disable active side TFO in certain scenarios

Middlebox firewall issues can potentially cause server's data being
blackholed after a successful 3WHS using TFO. Following are the related
reports from Apple:
https://www.nanog.org/sites/default/files/Paasch_Network_Support.pdf
Slide 31 identifies an issue where the client ACK to the server's data
sent during a TFO'd handshake is dropped.
C ---> syn-data ---> S
C <--- syn/ack ----- S
C (accept & write)
C <---- data ------- S
C ----- ACK -> X     S
[retry and timeout]

https://www.ietf.org/proceedings/94/slides/slides-94-tcpm-13.pdf
Slide 5 shows a similar situation that the server's data gets dropped
after 3WHS.
C ---- syn-data ---> S
C <--- syn/ack ----- S
C ---- ack --------> S
S (accept & write)
C?  X <- data ------ S
[retry and timeout]

This is the worst failure b/c the client can not detect such behavior to
mitigate the situation (such as disabling TFO). Failing to proceed, the
application (e.g., SSL library) may simply timeout and retry with TFO
again, and the process repeats indefinitely.

The proposed solution is to disable active TFO globally under the
following circumstances:
1. client side TFO socket detects out of order FIN
2. client side TFO socket receives out of order RST

We disable active side TFO globally for 1hr at first. Then if it
happens again, we disable it for 2h, then 4h, 8h, ...
And we reset the timeout to 1hr if a client side TFO sockets not opened
on loopback has successfully received data segs from server.
And we examine this condition during close().

The rational behind it is that when such firewall issue happens,
application running on the client should eventually close the socket as
it is not able to get the data it is expecting. Or application running
on the server should close the socket as it is not able to receive any
response from client.
In both cases, out of order FIN or RST will get received on the client
given that the firewall will not block them as no data are in those
frames.
And we want to disable active TFO globally as it helps if the middle box
is very close to the client and most of the connections are likely to
fail.

Also, add a debug sysctl:
  tcp_fastopen_blackhole_detect_timeout_sec:
    the initial timeout to use when firewall blackhole issue happens.
    This can be set and read.
    When setting it to 0, it means to disable the active disable logic.

Signed-off-by: Wei Wang <weiwan@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge tag 'mlx5-updates-2017-04-22' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Mon, 24 Apr 2017 18:11:10 +0000 (14:11 -0400)]
Merge tag 'mlx5-updates-2017-04-22' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2017-04-22

Sparse and compiler warnings fixes from Stephen Hemminger.

From Roi Dayan and Or Gerlitz, Add devlink and mlx5 support for controlling
E-Switch encapsulation mode, this knob will enable HW support for applying
encapsulation/decapsulation to VF traffic as part of SRIOV e-switch offloading.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: add rcu locking when changing early demux
David Ahern [Sat, 22 Apr 2017 16:33:16 +0000 (09:33 -0700)]
net: add rcu locking when changing early demux

systemd-sysctl is triggering a suspicious RCU usage message when
net.ipv4.tcp_early_demux or net.ipv4.udp_early_demux is changed via
a sysctl config file:

[   33.896184] ===============================
[   33.899558] [ ERR: suspicious RCU usage.  ]
[   33.900624] 4.11.0-rc7+ #104 Not tainted
[   33.901698] -------------------------------
[   33.903059] /home/dsa/kernel-2.git/net/ipv4/sysctl_net_ipv4.c:305 suspicious rcu_dereference_check() usage!
[   33.905724]
other info that might help us debug this:

[   33.907656]
rcu_scheduler_active = 2, debug_locks = 0
[   33.909288] 1 lock held by systemd-sysctl/143:
[   33.910373]  #0:  (sb_writers#5){.+.+.+}, at: [<ffffffff8123a370>] file_start_write+0x45/0x48
[   33.912407]
stack backtrace:
[   33.914018] CPU: 0 PID: 143 Comm: systemd-sysctl Not tainted 4.11.0-rc7+ #104
[   33.915631] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140531_083030-gandalf 04/01/2014
[   33.917870] Call Trace:
[   33.918431]  dump_stack+0x81/0xb6
[   33.919241]  lockdep_rcu_suspicious+0x10f/0x118
[   33.920263]  proc_configure_early_demux+0x65/0x10a
[   33.921391]  proc_udp_early_demux+0x3a/0x41

add rcu locking to proc_configure_early_demux.

Fixes: dddb64bcb3461 ("net: Add sysctl to toggle early demux for tcp and udp")
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'for-upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetoot...
David S. Miller [Mon, 24 Apr 2017 18:05:40 +0000 (14:05 -0400)]
Merge branch 'for-upstream' of git://git./linux/kernel/git/bluetooth/bluetooth-next

Johan Hedberg says:

====================
pull request: bluetooth-next 2017-04-22

Here are some more Bluetooth patches (and one 802.15.4 patch) in the
bluetooth-next tree targeting the 4.12 kernel. Most of them are pure
fixes.

Please let me know if there are any issues pulling. Thanks.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: netcp: fix spelling mistake: "memomry" -> "memory"
Colin Ian King [Sat, 22 Apr 2017 12:22:01 +0000 (13:22 +0100)]
net: netcp: fix spelling mistake: "memomry" -> "memory"

Trivial fix to spelling mistake in dev_err message and rejoin
line.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: atheros: atl1: use offset_in_page() macro
Geliang Tang [Sat, 22 Apr 2017 01:21:10 +0000 (09:21 +0800)]
net: atheros: atl1: use offset_in_page() macro

Use offset_in_page() macro instead of open-coding.

Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'bnxt_en-misc-next'
David S. Miller [Mon, 24 Apr 2017 17:54:48 +0000 (13:54 -0400)]
Merge branch 'bnxt_en-misc-next'

Michael Chan says:

====================
bnxt_en: Updates for net-next.

Miscellaneous updates include passing DCBX RoCE VLAN priority to firmware,
checking one more new firmware flag before allowing DCBX to run on the host,
adding 100Gbps speed support, adding check to disallow speed settings on
Multi-host NICs, and a minor fix for reporting VF attributes.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>