platform/kernel/linux-rpi.git
6 years agonet: Export open_related_ns()
Kirill Tkhai [Wed, 14 Feb 2018 13:40:05 +0000 (16:40 +0300)]
net: Export open_related_ns()

This function will be used to obtain net of tun device.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Make extern and export get_net_ns()
Kirill Tkhai [Wed, 14 Feb 2018 13:39:56 +0000 (16:39 +0300)]
net: Make extern and export get_net_ns()

This function will be used to obtain net of tun device.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Wed, 14 Feb 2018 20:44:05 +0000 (15:44 -0500)]
Merge branch '40GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
40GbE Intel Wired LAN Driver Updates 2018-02-14

This patch series enables the new mqprio hardware offload mechanism
creating traffic classes on VFs for XL710 devices. The parameters
needed to configure these traffic classes/queue channels are provides
by the user via the tc tool. A maximum of four traffic classes can be
created on each VF. This patch series also enables application of cloud
filters to each of these traffic classes. The cloud filters are applied
using the tc-flower classifier.

Example:
    1. tc qdisc add dev vf0 root mqprio num_tc 4 map 0 0 0 0 1 2 2 3\
        queues 2@0 2@2 1@4 1@5 hw 1 mode channel
    2. tc qdisc add dev vf0 ingress
    3. ethtool -K vf0 hw-tc-offload on
    4. ip link set eth0 vf 0 spoofchk off
    5. tc filter add dev vf0 protocol ip parent ffff: prio 1 flower dst_ip\
        192.168.3.5/32 ip_proto udp dst_port 25 skip_sw hw_tc 2
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agokcm: Call strp_stop before strp_done in kcm_attach
Tom Herbert [Wed, 14 Feb 2018 17:22:42 +0000 (09:22 -0800)]
kcm: Call strp_stop before strp_done in kcm_attach

In kcm_attach strp_done is called when sk_user_data is already
set to fail the attach. strp_done needs the strp to be stopped and
warns if it isn't. Call strp_stop in this case to eliminate the
warning message.

Reported-by: syzbot+88dfb55e4c8b770d86e3@syzkaller.appspotmail.com
Fixes: e5571240236c5652f ("kcm: Check if sk_user_data already set in kcm_attach"
Signed-off-by: Tom Herbert <tom@quantonium.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: dp83867: Add documentation for CLK_OUT pin muxing
Wadim Egorov [Wed, 14 Feb 2018 16:07:12 +0000 (17:07 +0100)]
net: phy: dp83867: Add documentation for CLK_OUT pin muxing

Add documentation of ti,clk-output-sel which can be used to select
a specific clock for CLK_OUT.

Signed-off-by: Wadim Egorov <w.egorov@phytec.de>
Signed-off-by: Daniel Schultz <d.schultz@phytec.de>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: dp83867: Add binding for the CLK_OUT pin muxing option
Wadim Egorov [Wed, 14 Feb 2018 16:07:11 +0000 (17:07 +0100)]
net: phy: dp83867: Add binding for the CLK_OUT pin muxing option

The DP83867 has a muxing option for the CLK_OUT pin. It is possible
to set CLK_OUT for different channels.
Create a binding to select a specific clock for CLK_OUT pin.

Signed-off-by: Wadim Egorov <w.egorov@phytec.de>
Signed-off-by: Daniel Schultz <d.schultz@phytec.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotipc: apply bearer link tolerance on running links
Jon Maloy [Wed, 14 Feb 2018 12:34:39 +0000 (13:34 +0100)]
tipc: apply bearer link tolerance on running links

Currently, the default link tolerance set in struct tipc_bearer only
has effect on links going up after that moment. I.e., a user has to
reset all the node's links across that bearer to have the new value
applied. This is too limiting and disturbing on a running cluster to
be useful.

We now change this so that also already existing links are updated
dynamically, without any need for a reset, when the bearer value is
changed. We leverage the already existing per-link functionality
for this to achieve the wanted effect.

Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'cxgb4-speed-up-reading-on-chip-memory'
David S. Miller [Wed, 14 Feb 2018 20:01:52 +0000 (15:01 -0500)]
Merge branch 'cxgb4-speed-up-reading-on-chip-memory'

Rahul Lakkireddy says:

====================
cxgb4: speed up reading on-chip memory

This series of patches speed up reading on-chip memory (EDC and MC)
by reading 64-bits at a time.

Patch 1 reworks logic to read EDC and MC.

Patch 2 adds logic to read EDC and MC 64-bits at a time.

v2:
- Dropped AVX CPU intrinsic instructions.
- Use readq() to read 64-bits at a time.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocxgb4: speed up on-chip memory read
Rahul Lakkireddy [Wed, 14 Feb 2018 07:26:28 +0000 (12:56 +0530)]
cxgb4: speed up on-chip memory read

Use readq() (via t4_read_reg64()) to read 64-bits at a time.
Read residual in 32-bit multiples.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocxgb4: rework on-chip memory read
Rahul Lakkireddy [Wed, 14 Feb 2018 07:26:27 +0000 (12:56 +0530)]
cxgb4: rework on-chip memory read

Rework logic to read EDC and MC. Do 32-bit reads at a time.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Move ipv4 set_lwt_redirect helper to lwtunnel
David Ahern [Wed, 14 Feb 2018 04:32:04 +0000 (20:32 -0800)]
net: Move ipv4 set_lwt_redirect helper to lwtunnel

IPv4 uses set_lwt_redirect to set the lwtunnel redirect functions as
needed. Move it to lwtunnel.h as lwtunnel_set_redirect and change
IPv6 to also use it.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'PTP-support-for-DSA-and-mv88e6xxx-driver'
David S. Miller [Wed, 14 Feb 2018 19:33:38 +0000 (14:33 -0500)]
Merge branch 'PTP-support-for-DSA-and-mv88e6xxx-driver'

Andrew Lunn says:

====================
PTP support for DSA and mv88e6xxx driver.

This patchset adds support for using the PTP hardware in switches
supported by the mv88e6xxx driver. The code was produces in
collaboration with Brandon Streiff doing the initial implementation,
and then Richard Cochran and Andrew Lunn making further changes and
cleanups.

The code is sufficient to use ptp4l on a single DSA interface, either
as a master or a slave. Due to the use of an MDIO bus to access the
switch, reading hardware timestamps is slower than what ptp4l
expects. Thus it is necessary to use the option
--tx_timestamp_timeout=32. Heavy use of ethtool -S, or bridge fdb show
can also upset ptp4l. Patches to address this will follow.

Further work is requires to support bridges using Boundary Clock or
Transparent Clock mode.

Since the RFC, an overflow bug has been fixed. Brandon Streiff
has also Acked-by: the updates to his initial patchset.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: mv88e6xxx: add workaround for 6341 timestamping
Brandon Streiff [Wed, 14 Feb 2018 00:07:51 +0000 (01:07 +0100)]
net: dsa: mv88e6xxx: add workaround for 6341 timestamping

88E6341 devices default to timestamping at the PHY, but due to a
hardware issue, timestamps via this component are unreliable. For
this family, configure the PTP hardware to force the timestamping
to occur at the MAC.

Signed-off-by: Brandon Streiff <brandon.streiff@ni.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: mv88e6xxx: add rx/tx timestamping support
Brandon Streiff [Wed, 14 Feb 2018 00:07:50 +0000 (01:07 +0100)]
net: dsa: mv88e6xxx: add rx/tx timestamping support

This patch implements RX/TX timestamping support.

The Marvell PTP hardware supports RX timestamping individual message
types, but for simplicity we only support the EVENT receive filter since
few if any clients bother with the more specific filter types.

checkpatch and reverse Christmas tree changes by Andrew Lunn.

Re-factor duplicated code paths and avoid IfOk anti-pattern, use the
common ptp worker thread from the class layer and time stamp UDP/IPv4
frames as well as Layer-2 frame by Richard Cochran.

Signed-off-by: Brandon Streiff <brandon.streiff@ni.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: forward timestamping callbacks to switch drivers
Brandon Streiff [Wed, 14 Feb 2018 00:07:49 +0000 (01:07 +0100)]
net: dsa: forward timestamping callbacks to switch drivers

Forward the rx/tx timestamp machinery from the dsa infrastructure to the
switch driver.

On the rx side, defer delivery of skbs until we have an rx timestamp.
This mimicks the behavior of skb_defer_rx_timestamp.

On the tx side, identify PTP packets, clone them, and pass them to the
underlying switch driver before we transmit. This mimicks the behavior
of skb_tx_timestamp.

Adjusted txstamp API to keep the allocation and freeing of the clone
in the same central function by Richard Cochran

Signed-off-by: Brandon Streiff <brandon.streiff@ni.com>
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: forward hardware timestamping ioctls to switch driver
Brandon Streiff [Wed, 14 Feb 2018 00:07:48 +0000 (01:07 +0100)]
net: dsa: forward hardware timestamping ioctls to switch driver

This patch adds support to the dsa slave network device so that
switch drivers can implement the SIOC[GS]HWTSTAMP ioctls and the
ethtool timestamp-info interface.

Signed-off-by: Brandon Streiff <brandon.streiff@ni.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: mv88e6xxx: add support for event capture
Brandon Streiff [Wed, 14 Feb 2018 00:07:47 +0000 (01:07 +0100)]
net: dsa: mv88e6xxx: add support for event capture

This patch adds support for configuring mv88e6xxx GPIO lines as PTP
pins, so that they may be used for time stamping external events or for
periodic output.

Checkpatch and reverse Christmas tree fixes by Andrew Lunn

Periodic output removed by Richard Cochran, until a better abstraction
of a VCO is added to Linux in general.

Signed-off-by: Brandon Streiff <brandon.streiff@ni.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: mv88e6xxx: add support for GPIO configuration
Brandon Streiff [Wed, 14 Feb 2018 00:07:46 +0000 (01:07 +0100)]
net: dsa: mv88e6xxx: add support for GPIO configuration

MV88E6352 and later switches support GPIO control through the "Scratch
& Misc" global2 register. (Older switches do too, though with a slightly
different register interface. Only the 6352-style is implemented here.)

Add a new file, global2_scratch.c, for operations in the Scratch & Misc
space. Additionally, add a GPIO operations structure to present an
abstract view over GPIO manipulation.

Reverse Christmas tree and unsigned has been replaced with unsigned
int by Andrew Lunn.

Signed-off-by: Brandon Streiff <brandon.streiff@ni.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: mv88e6xxx: expose switch time as a PTP hardware clock
Brandon Streiff [Wed, 14 Feb 2018 00:07:45 +0000 (01:07 +0100)]
net: dsa: mv88e6xxx: expose switch time as a PTP hardware clock

This patch adds basic support for exposing the 32-bit timestamp counter
inside the mv88e6xxx switch as a ptp_clock.

Adjfine implemented by Richard Cochran.
Andrew Lunn: fix return value of PTP stub function.

Signed-off-by: Brandon Streiff <brandon.streiff@ni.com>
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: mv88e6xxx: add accessors for PTP/TAI registers
Brandon Streiff [Wed, 14 Feb 2018 00:07:44 +0000 (01:07 +0100)]
net: dsa: mv88e6xxx: add accessors for PTP/TAI registers

This patch implements support for accessing the Precision Time Protocol
and Time Application Interface registers via the AVB register interface
in the Global 2 register.

The register interface differs slightly between different models; older
models use a 3-bit operations field, while newer models use a 2-bit
field. The operations values and the special "global port" values are
different between the two. This is a similar split to the differences
in the "Ingress Rate" register between models, so, like in that case,
we call the two variants "6352" and "6390" and create an ops structure
to abstract between the two.

checkpatch fixups by Andrew Lunn

Signed-off-by: Brandon Streiff <brandon.streiff@ni.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: mv88e6xxx: export g2 register accessors
Brandon Streiff [Wed, 14 Feb 2018 00:07:43 +0000 (01:07 +0100)]
net: dsa: mv88e6xxx: export g2 register accessors

Let the mv88e6xxx_g2_* register accessor functions be accessible
outside of global2.c.

Signed-off-by: Brandon Streiff <brandon.streiff@ni.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ptp: Add stub for ptp_classify_raw()
Andrew Lunn [Wed, 14 Feb 2018 00:07:42 +0000 (01:07 +0100)]
net: ptp: Add stub for ptp_classify_raw()

When NET_PTP_CLASSIFY is disabled, a stub function is required in
order that the drivers compile.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: try to keep packet if SYN_RCV race is lost
Eric Dumazet [Tue, 13 Feb 2018 14:14:12 +0000 (06:14 -0800)]
tcp: try to keep packet if SYN_RCV race is lost

배석진 reported that in some situations, packets for a given 5-tuple
end up being processed by different CPUS.

This involves RPS, and fragmentation.

배석진 is seeing packet drops when a SYN_RECV request socket is
moved into ESTABLISH state. Other states are protected by socket lock.

This is caused by a CPU losing the race, and simply not caring enough.

Since this seems to occur frequently, we can do better and perform
a second lookup.

Note that all needed memory barriers are already in the existing code,
thanks to the spin_lock()/spin_unlock() pair in inet_ehash_insert()
and reqsk_put(). The second lookup must find the new socket,
unless it has already been accepted and closed by another cpu.

Note that the fragmentation could be avoided in the first place by
use of a correct TCP MSS option in the SYN{ACK} packet, but this
does not mean we can not be more robust.

Many thanks to 배석진 for a very detailed analysis.

Reported-by: 배석진 <soukjin.bae@samsung.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoi40e: Add and delete cloud filter
Avinash Dayanand [Tue, 23 Jan 2018 16:51:06 +0000 (08:51 -0800)]
i40e: Add and delete cloud filter

This patch provides support to add or delete cloud filter for queue
channels created for ADq on VF.
We are using the HW's cloud filter feature and programming it to act
as a TC filter applied to a group of queues.

There are two possible modes for a VF when applying a cloud filter
1. Basic Mode: Intended to apply filters that don't need a VF to be
Trusted. This would include the following
  Dest MAC + L4 port
  Dest MAC + VLAN + L4 port
2. Advanced Mode: This mode is only for filters with combination that
  requires VF to be Trusted.
  Dest IP + L4 port

When cloud filters are applied on a trusted VF and for some reason
the same VF is later made as untrusted then all cloud filters
will be deleted. All cloud filters has to be re-applied in
such a case.
Cloud filters are also deleted when queue channel is deleted.

Testing-Hints:
=============
1. Adding Basic Mode filter should be possible on a VF in
   Non-Trusted mode.
2. In Advanced mode all filters should be able to be created.

Steps:
======
1. Enable ADq and create TCs using TC mqprio command
2. Apply cloud filter.
3. Turn-off the spoof check.
4. Pass traffic.

Example:
========
1. tc qdisc add dev enp4s2 root mqprio num_tc 4 map 0 0 0 0 1 2 2 3\
queues 2@0 2@2 1@4 1@5 hw 1 mode channel
2. tc qdisc add dev enp4s2 ingress
3. ethtool -K enp4s2 hw-tc-offload on
4. ip link set ens261f0 vf 0 spoofchk off
5. tc filter add dev enp4s2 protocol ip parent ffff: prio 1 flower\
dst_ip 192.168.3.5/32 ip_proto udp dst_port 25 skip_sw hw_tc 2

Signed-off-by: Avinash Dayanand <avinash.dayanand@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40evf: Add support to apply cloud filters
Harshitha Ramamurthy [Tue, 23 Jan 2018 16:51:05 +0000 (08:51 -0800)]
i40evf: Add support to apply cloud filters

This patch enables a tc filter to be applied as a cloud
filter for the VF. This patch adds functions which parse the
tc filter, extract the necessary fields needed to configure the
filter and package them in a virtchnl message to be sent to the
PF to apply them.

Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agovirtchnl: Add filter data structures
Harshitha Ramamurthy [Tue, 23 Jan 2018 16:51:04 +0000 (08:51 -0800)]
virtchnl: Add filter data structures

This patch adds infrastructure to send virtchnl messages to the
PF to configure filters on the VF. The patch adds a struct
called virtchnl_filter which contains information about the fields
in the user-specified tc filter.

Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agovirtchnl: Add a macro to check the size of a union
Harshitha Ramamurthy [Tue, 23 Jan 2018 16:51:03 +0000 (08:51 -0800)]
virtchnl: Add a macro to check the size of a union

This patch adds a macro to check if the size of a union is correct.
It throws a divide by zero error if the union is not of the correct
size.

Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40e: Service request to configure bandwidth for ADq on a VF
Avinash Dayanand [Tue, 23 Jan 2018 16:51:02 +0000 (08:51 -0800)]
i40e: Service request to configure bandwidth for ADq on a VF

This patch handles the request from ADq enabled VF to allocate
bandwidth to each traffic class which means for each VSI.

Signed-off-by: Avinash Dayanand <avinash.dayanand@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40evf: Add support to configure bw via tc tool
Harshitha Ramamurthy [Tue, 23 Jan 2018 16:51:01 +0000 (08:51 -0800)]
i40evf: Add support to configure bw via tc tool

This patch adds support to configure bandwidth for the traffic
classes via tc tool. The required information is passed to the PF
which is used in the process of setting up the traffic classes.

Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40e: Delete queue channel for ADq on VF
Avinash Dayanand [Tue, 23 Jan 2018 16:51:00 +0000 (08:51 -0800)]
i40e: Delete queue channel for ADq on VF

This patch takes care of freeing up all the VSIs, queues and
other ADq related software and hardware resources, when a user
requests for deletion of ADq on VF.

Example command:
tc qdisc del dev eth0 root

Signed-off-by: Avinash Dayanand <avinash.dayanand@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40evf: Alloc queues for ADq on VF
Avinash Dayanand [Tue, 23 Jan 2018 16:50:59 +0000 (08:50 -0800)]
i40evf: Alloc queues for ADq on VF

This patch allocates number of queues requested by the user as a part
of TC command when ADq is enabled on a VF.

In order to be consistent in design with PF implementation of ADq,
don't allow to set channels via ethtool from VF when ADq is already
enabled. This means the users will not be able to change the number of
queues/channels via ethtool for a VF when ADq is ON. In order to be
able to use set channels, users will be required to disable ADq first
and then try setting the channels again.

When ADq is enabled on VF, it goes through a reset during which VSIs
and queues are re-configured. Meanwhile if we receive link status
message from PF even before the queues are re-configured, just ignore
this link up message.

Signed-off-by: Avinash Dayanand <avinash.dayanand@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40e: Enable ADq and create queue channel/s on VF
Avinash Dayanand [Tue, 23 Jan 2018 16:50:58 +0000 (08:50 -0800)]
i40e: Enable ADq and create queue channel/s on VF

This patch enables ADq and creates queue channels on a VF. An ADq
enabled VF can have up to 4 VSIs and each one of them represents
a traffic class and this is termed as a queue channel. Each of these
VSIs can have up to 4 queues. This patch services the request for
enabling ADq and adds queue channel based on the TC mqprio info
provided by the user in the VF.

Initially a check is made to see if spoof check is OFF, if not ADq
will not be enabled. PF notifies VF for a reset in order to complete
the creation of ADq resources i.e. creation of additional VSIs and
allocation of queues as per TC information, all in the reset path.

Steps:
======
1. Turn off the spoof check
2. Enable ADq using tc mqprio command with or without rate limit.
3. Pass traffic.

Example:
========
% ip link set dev eth0 vf 0 spoofchk off
% tc qdisc add dev $iface root mqprio num_tc 4 map\
0 0 0 0 1 1 1 1 2 2 2 2 3 3 3 3 queues\
4@0 4@4 4@8 4@8 hw 1 mode channel

Expected results:
=================
1. Total number of queues for the VF should be sum of queues of all TCs.
2. Traffic flow should be normal without errors.

Signed-off-by: Avinash Dayanand <avinash.dayanand@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40evf: add ndo_setup_tc callback to i40evf
Harshitha Ramamurthy [Tue, 23 Jan 2018 16:50:57 +0000 (08:50 -0800)]
i40evf: add ndo_setup_tc callback to i40evf

This patch introduces the callback to the ndo_setup_tc function
in the VF driver. We add a wrapper function to make room for the
upcoming cloud filter patches which add calls to different functions
from setup_tc.

First, we add support for capability exchange for ADQ between the
PF and VF. Next, we add support to take in the mqprio configuration
and configure queues as per the traffic classes, rate limit and the
priorities specified by the user. This is done by passing the channel
config to the PF driver through a virtchannel message.

The flags and bits added, track if ADq is enabled, set max number of
traffic classes to 4 and provide ability to negotiate capability with
the PF.

Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agovirtchnl: Add virtchl structures to support queue channels
Harshitha Ramamurthy [Tue, 23 Jan 2018 16:50:56 +0000 (08:50 -0800)]
virtchnl: Add virtchl structures to support queue channels

This patch defines new structs in support of the virtchannel message
that the VF sends to the PF to create a queue channel specified by the
user via tc tool.

Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40evf: Fix link up issue when queues are disabled
Avinash Dayanand [Tue, 23 Jan 2018 16:50:55 +0000 (08:50 -0800)]
i40evf: Fix link up issue when queues are disabled

One of the previous patch fixes the link up issue by ignoring it if
i40evf is not in __I40EVF_RUNNING state. However this doesn't fix the
race condition when queues are disabled esp for ADq on VF. Hence check
if all queues are enabled before starting all queues.

Signed-off-by: Avinash Dayanand <avinash.dayanand@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoMerge branch 'net-dev-Make-protocol-ptr-dependent-on-CONFIG'
David S. Miller [Wed, 14 Feb 2018 16:55:39 +0000 (11:55 -0500)]
Merge branch 'net-dev-Make-protocol-ptr-dependent-on-CONFIG'

David Ahern says:

====================
net: dev: Make protocol ptr dependent on CONFIG

Found these in a branch from 3-years ago. Still relevant today.
Make decnet, ax25, and atalk ptrs in net_device based on their
respective CONFIG.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Remove atalk header from socket.c
David Ahern [Tue, 13 Feb 2018 16:52:03 +0000 (08:52 -0800)]
net: Remove atalk header from socket.c

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Make atalk_ptr depend on ATALK or IRDA
David Ahern [Tue, 13 Feb 2018 16:52:02 +0000 (08:52 -0800)]
net: Make atalk_ptr depend on ATALK or IRDA

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Make ax25_ptr depend on CONFIG_AX25
David Ahern [Tue, 13 Feb 2018 16:52:01 +0000 (08:52 -0800)]
net: Make ax25_ptr depend on CONFIG_AX25

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Make dn_ptr depend on CONFIG_DECNET
David Ahern [Tue, 13 Feb 2018 16:52:00 +0000 (08:52 -0800)]
net: Make dn_ptr depend on CONFIG_DECNET

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Wed, 14 Feb 2018 03:31:23 +0000 (22:31 -0500)]
Merge branch '40GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
40GbE Intel Wired LAN Driver Updates 2018-02-13

This series contains updates to i40e and i40evf.

Wei Yongjun fixes a function that needed to be "static".  Also fixes the
use of GFP_KERNEL to GFP_ATOMIC when we have taken a spinlock.

Mitch cleans up several info messages to not include the memory
addresses being used on the off chance this information could be used
maliciously.

Alan provides several fixes to the broadcast filters starting with the
triggering of overflow promiscuous in circumstances where we run out of
space for broadcast filters to prevent traffic from being unexpectedly
dropped.  Refactored the code to improve the readability and
maintainability when we are concerned about when and how overflow
promiscuous is changed.

Harshitha cleans up a message to make it more clear on what is being
reset, so users are not confused and think the PF is resetting.

Dave fixes an issue where the MAC, firmware version and NPAR checks used
to determine if shutting off the firmware LLDP engine is supported or
not, instead set a hardware flag which ethtool can use.

Jake updates the VF driver to use __dev_uc_sync and __dev_mc_sync, like
the PF driver.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: Add FIB onlink tests
David Ahern [Tue, 13 Feb 2018 16:44:06 +0000 (08:44 -0800)]
selftests: Add FIB onlink tests

Add test cases verifying FIB onlink commands work as expected in
various conditions - IPv4, IPv6, main table, and VRF.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoi40evf: Make VF reset warning message more clear
Harshitha Ramamurthy [Mon, 22 Jan 2018 17:00:39 +0000 (12:00 -0500)]
i40evf: Make VF reset warning message more clear

When the PF resets the VF, the VF puts out a warning message
indicating that the VF received a reset message from the PF.
Make this message more clear so that we do not mistakenly
think that the PF is undergoing a reset.

Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40evf: use __dev_[um]c_sync routines in .set_rx_mode
Jacob Keller [Mon, 22 Jan 2018 17:00:38 +0000 (12:00 -0500)]
i40evf: use __dev_[um]c_sync routines in .set_rx_mode

Similar to changes done to the PF driver in commit 6622f5cdbaf3 ("i40e:
make use of __dev_uc_sync and __dev_mc_sync"), replace our
home-rolled method for updating the internal status of MAC filters with
__dev_uc_sync and __dev_mc_sync.

These new functions use internal state within the netdev struct in order
to efficiently break the question of "which filters in this list need to
be added or removed" into singular "add this filter" and "delete this
filter" requests.

This vastly improves our handling of .set_rx_mode especially with large
number of MAC filters being added to the device, and even results in
a simpler .set_rx_mode handler.

Under some circumstances, such as when attached to a bridge, we may
receive a request to delete our own permanent address. Prevent deletion
of this address during i40evf_addr_unsync so that we don't accidentally
stop receiving traffic.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40e: i40e: Change ethtool check from MAC to HW flag
Dave Ertman [Mon, 22 Jan 2018 17:00:37 +0000 (12:00 -0500)]
i40e: i40e: Change ethtool check from MAC to HW flag

The MAC, FW Version and NPAR check used to determine
if shutting off the FW LLDP engine is supported is not
using the usual feature check mechanism.

This patch fixes the problem by moving the feature check
to i40e_sw_init in order to set a flag in pf->hw_features
that ethtool will use for priv_flags disable operation.

Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40e: do not force filter failure in overflow promiscuous
Alan Brady [Mon, 22 Jan 2018 17:00:36 +0000 (12:00 -0500)]
i40e: do not force filter failure in overflow promiscuous

Broadcast filters can now cause overflow promiscuous to trigger when
adding "too many" VLANs to all the ports of a device and the driver
needs a way to exit overflow promiscuous once triggered.

Currently the driver looks to see if there are "too many" filters and/or
we have any failed filters to determine when it is safe to exit overflow
promiscuous.  If we trigger overflow promiscuous with broadcast filters,
any new filters added will be "auto-failed" until we exit overflow
promiscuous.  Since the user can't manually remove the failed broadcast
filters for VLANs (nor should we expect the user to do such), there is
no way to exit overflow promiscuous without reloading the driver.

The easiest way to do this is to remove the shortcut to "auto-fail"
filters in overflow promiscuous.  If the user removes the VLANs, the
failed filters will be removed and since we're no longer "auto-failing"
new filters, we'll eventually get a good set of filters and exit
overflow promiscuous.

This has the side benefit of making filter state more explicit in that
if a filter says it's failed we know for a fact it failed and not just
assuming it will if we're in overflow promiscuous.  This is nice because
if the user removes some filters and then adds some, even if we're in
overflow promiscuous, the filter might succeed; we were just assuming it
won't because the user hasn't rectified other existing failed filters.

Signed-off-by: Alan Brady <alan.brady@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40e: refactor promisc_changed in i40e_sync_vsi_filters
Alan Brady [Mon, 22 Jan 2018 17:00:35 +0000 (12:00 -0500)]
i40e: refactor promisc_changed in i40e_sync_vsi_filters

This code here is quite complex and easy to screw up.  Let's see if we
can't improve the readability and maintainability a bit.  This refactors
out promisc_changed into two variables 'old_overflow' and 'new_overflow'
which makes it a bit clearer when we're concerned about when and how
overflow promiscuous is changed.  This also makes so that we no longer
need to pass a boolean pointer to i40e_aqc_add_filters.  Instead we can
simply check if we changed the overflow promiscuous flag since the
function start.

Signed-off-by: Alan Brady <alan.brady@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40evf: Use an iterator of the same type as the list
Harshitha Ramamurthy [Mon, 22 Jan 2018 17:00:34 +0000 (12:00 -0500)]
i40evf: Use an iterator of the same type as the list

When iterating through the linked list of VLAN filters, make the
iterator the same type as that of the linked list.

Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40e: broadcast filters can trigger overflow promiscuous
Alan Brady [Mon, 22 Jan 2018 17:00:33 +0000 (12:00 -0500)]
i40e: broadcast filters can trigger overflow promiscuous

When adding a bunch of VLANs to all the ports on a device, it's possible
to run out of space for broadcast filters.  The driver should trigger
overflow promiscuous in this circumstance to prevent traffic from being
unexpectedly dropped.

Signed-off-by: Alan Brady <alan.brady@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40e: don't leak memory addresses
Mitch Williams [Mon, 22 Jan 2018 17:00:32 +0000 (12:00 -0500)]
i40e: don't leak memory addresses

Could a Bad Person do Bad Things to a server if they found these
addresses printed in the log? Who knows? But let's not take that risk.

Remove pointers from a bunch of printks. In some cases, I was able to
adjust the message to indicate whether or not the value was null. In
others, I just removed the entire message as there was really no hope of
saving it.

Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40evf: use GFP_ATOMIC under spin lock
Wei Yongjun [Fri, 12 Jan 2018 02:27:13 +0000 (02:27 +0000)]
i40evf: use GFP_ATOMIC under spin lock

A spin lock is taken here so we should use GFP_ATOMIC.

Fixes: 504398f0a78e ("i40evf: use spinlock to protect (mac|vlan)_filter_list")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Acked-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoi40e: Make local function i40e_get_link_speed static
Wei Yongjun [Wed, 10 Jan 2018 07:13:51 +0000 (07:13 +0000)]
i40e: Make local function i40e_get_link_speed static

Fixes the following sparse warning:

drivers/net/ethernet/intel/i40e/i40e_main.c:5440:5: warning:
 symbol 'i40e_get_link_speed' was not declared. Should it be static?

Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoMerge branch 'selftests-fib_tests-simplifications-verbosity-and-a-race'
David S. Miller [Tue, 13 Feb 2018 19:34:29 +0000 (14:34 -0500)]
Merge branch 'selftests-fib_tests-simplifications-verbosity-and-a-race'

David Ahern says:

====================
selftests: fib_tests: simplifications, verbosity and a race

Improve efficiency of fib_tests.sh and make the test result more verbose,
from this summary:
$ fib_tests.sh is failing in a VM:
    $ fib_tests.sh
    Running netdev unregister tests
    PASS: unicast route test
    PASS: multipath route test
    Running netdev down tests
    PASS: unicast route test
    PASS: multipath route test
    Running netdev carrier change tests
    PASS: local route carrier test
    FAIL: unicast route carrier test

where a single entry actually corresponds to many checks to a much more
verbse output that clarifies test cases:
$fib_tests.sh
Single path route carrier test
    ....
    Carrier down
        IPv4 fibmatch                                         [ OK ]
        IPv6 fibmatch                                         [ OK ]
        IPv4 linkdown flag set                                [FAIL]
        IPv6 linkdown flag set                                [FAIL]
    Second address added with carrier down
        IPv4 fibmatch                                         [ OK ]
        IPv6 fibmatch                                         [ OK ]
        IPv4 linkdown flag set                                [FAIL]
        IPv6 linkdown flag set                                [ OK ]

And then fix the race in changing carrier down on dummy device to checking
the corresponding routes.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: fib_tests: sleep after changing carrier
David Ahern [Tue, 13 Feb 2018 16:37:36 +0000 (08:37 -0800)]
selftests: fib_tests: sleep after changing carrier

sleep for a second after setting carrier down to allow linkwatch
to propagate the change to the routing stack via netdev_state_change.
As it stands there is a race setting carrier down on the dummy
device and then checking the linkdown flag in the routes.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: fib_tests: Move admin of dummy0 to helpers
David Ahern [Tue, 13 Feb 2018 16:37:35 +0000 (08:37 -0800)]
selftests: fib_tests: Move admin of dummy0 to helpers

Move setup and teardown of testns and dummy0 to helpers.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: fib_tests: Make test results more verbose
David Ahern [Tue, 13 Feb 2018 16:37:34 +0000 (08:37 -0800)]
selftests: fib_tests: Make test results more verbose

fib_tests.sh is failing in a VM:
    $ fib_tests.sh
    Running netdev unregister tests
    PASS: unicast route test
    PASS: multipath route test
    Running netdev down tests
    PASS: unicast route test
    PASS: multipath route test
    Running netdev carrier change tests
    PASS: local route carrier test
    FAIL: unicast route carrier test

The last test corresponds to fib_carrier_unicast_test which 12 places
that could be failing. Be more verbose in the output so a failure is
easier to track down and separate test setup failures with set -e and
set +e pairs.

With the verbose logging it is easier to see which checks are failing:
    $fib_tests.sh
    Single path route carrier test
        ....
        Carrier down
            IPv4 fibmatch                                         [ OK ]
            IPv6 fibmatch                                         [ OK ]
            IPv4 linkdown flag set                                [FAIL]
            IPv6 linkdown flag set                                [FAIL]
        Second address added with carrier down
            IPv4 fibmatch                                         [ OK ]
            IPv6 fibmatch                                         [ OK ]
            IPv4 linkdown flag set                                [FAIL]
            IPv6 linkdown flag set                                [ OK ]

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: fib_tests: simplify ip commands in a namespace
David Ahern [Tue, 13 Feb 2018 16:37:33 +0000 (08:37 -0800)]
selftests: fib_tests: simplify ip commands in a namespace

'ip netns exec testns ip' is more efficiently handled using 'ip -netns';
runs the ip command after switching the namespace and avoids an exec.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv4: Unexport fib_multipath_hash and fib_select_path
David Ahern [Tue, 13 Feb 2018 16:12:12 +0000 (08:12 -0800)]
net/ipv4: Unexport fib_multipath_hash and fib_select_path

Do not export fib_multipath_hash or fib_select_path; both are only used
by core ipv4 code.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv4: Simplify fib_select_path
David Ahern [Tue, 13 Feb 2018 16:11:34 +0000 (08:11 -0800)]
net/ipv4: Simplify fib_select_path

If flow oif is set and it is not an l3mdev, then fib_select_path
can jump to the source address check.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'sctp-rename-sctp-diag-file-and-add-file-comments-for-it'
David S. Miller [Tue, 13 Feb 2018 18:56:32 +0000 (13:56 -0500)]
Merge branch 'sctp-rename-sctp-diag-file-and-add-file-comments-for-it'

Xin Long says:

====================
sctp: rename sctp diag file and add file comments for it

This patchset is to remove the sctp_ prefix for sctp diag file,
and also to add the missing file comments for it.

v1->v2:
  split them into two patches as Marcelo suggested.
====================

Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: add file comments in diag.c
Xin Long [Tue, 13 Feb 2018 11:33:21 +0000 (19:33 +0800)]
sctp: add file comments in diag.c

This patch is to add the missing file comments for sctp diag file.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: rename sctp_diag.c as diag.c
Xin Long [Tue, 13 Feb 2018 11:33:20 +0000 (19:33 +0800)]
sctp: rename sctp_diag.c as diag.c

Remove 'sctp_' prefix for diag file, to keep consistent with other
files' names.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum: Use NL_SET_ERR_MSG_MOD
Arkadi Sharshevsky [Tue, 13 Feb 2018 10:29:05 +0000 (11:29 +0100)]
mlxsw: spectrum: Use NL_SET_ERR_MSG_MOD

Use NL_SET_ERR_MSG_MOD helper which adds the module name instead
of specifying the prefix each time.

Signed-off-by: Arkadi Sharshevsky <arkadis@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'mlxsw-SPAN-cleanups'
David S. Miller [Tue, 13 Feb 2018 17:26:26 +0000 (12:26 -0500)]
Merge branch 'mlxsw-SPAN-cleanups'

Jiri Pirko says:

====================
mlxsw: SPAN cleanups

In patch one of this short series, a misplaced pointer star is moved to
the correct place.

In the second patch, we observe that if SPAN entries carry their
reference count anyway, it's redundant to also carry a "used" flag.

In the third patch, SPAN support code is moved to a separate module.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum: Move SPAN code to separate module
Petr Machata [Tue, 13 Feb 2018 10:27:48 +0000 (11:27 +0100)]
mlxsw: spectrum: Move SPAN code to separate module

For the upcoming work on SPAN, it makes sense to move the current code
to a module of its own. It already has a well-defined API boundary to
the mirror management (which is used from matchall and ACL code). A
couple more functions need to be exported for the functions that
spectrum.c needs to use for MTU handling and subsystem init/fini.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum: Drop struct span_entry.used
Petr Machata [Tue, 13 Feb 2018 10:27:47 +0000 (11:27 +0100)]
mlxsw: spectrum: Drop struct span_entry.used

The member ref_count already determines whether a given SPAN entry is
used, and is as easy to use as a dedicated boolean.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum: Fix a coding style nit
Petr Machata [Tue, 13 Feb 2018 10:27:46 +0000 (11:27 +0100)]
mlxsw: spectrum: Fix a coding style nit

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'mlxsw-IPIP-cleanups'
David S. Miller [Tue, 13 Feb 2018 17:24:28 +0000 (12:24 -0500)]
Merge branch 'mlxsw-IPIP-cleanups'

Jiri Pirko says:

====================
mlxsw: IPIP cleanups

In the first patch, a forgotten #include is added. Even though the code
compiles as-is, the include is necessary for modules that should include
spectrum_ipip.h.

The second patch corrects an assumption that IPv6 tunnels use struct
ip_tunnel_parm to store tunnel parameters.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum: Distinguish between IPv4/6 tunnels
Petr Machata [Tue, 13 Feb 2018 10:26:09 +0000 (11:26 +0100)]
mlxsw: spectrum: Distinguish between IPv4/6 tunnels

struct ip_tunnel_parm, where GRE and several other tunnel types hold
information, is IPv4-specific. The current router / ipip code in mlxsw
however uses it as if it were generic.

Make it clear that it's not. Rename many functions from _params_ to
_params4_. mlxsw_sp_ipip_parms_saddr() and _daddr() take a proto
argument to dispatch on it. Move the dispatch logic to
mlxsw_sp_ipip_netdev_saddr() and _daddr(), and replace with
single-protocol functions.

In struct mlxsw_sp_ipip_entry, move the "parms" field to a (for the time
being, singleton) union. Update users throughout.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum_ipip: Add a forgotten include
Petr Machata [Tue, 13 Feb 2018 10:26:08 +0000 (11:26 +0100)]
mlxsw: spectrum_ipip: Add a forgotten include

struct ip_tunnel_parm, which is used in spectrum_ipip.h, is defined in
if_tunnel.h. However, the former neglects to include the latter.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodpaa_eth: fix incorrect comment
Jake Moroni [Tue, 13 Feb 2018 00:23:43 +0000 (19:23 -0500)]
dpaa_eth: fix incorrect comment

The comment stated that a thread was started, but
that is not the case.

Signed-off-by: Jake Moroni <mail@jakemoroni.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'Replacing-net_mutex-with-rw_semaphore'
David S. Miller [Tue, 13 Feb 2018 15:36:10 +0000 (10:36 -0500)]
Merge branch 'Replacing-net_mutex-with-rw_semaphore'

Kirill Tkhai says:

====================
Replacing net_mutex with rw_semaphore

this is the third version of the patchset introducing net_sem
instead of net_mutex. The patchset adds net_sem in addition
to net_mutex and allows pernet_operations to be "async". This
flag means, the pernet_operations methods are safe to be
executed with any other pernet_operations (un)initializing
another net.

If there are only async pernet_operations in the system,
net_mutex is not used either for setup_net() or for cleanup_net().

The pernet_operations converted in this patchset allow
to create minimal .config to have network working, and
the changes improve the performance like you may see
below:

    %for i in {1..10000}; do unshare -n bash -c exit; done

    *before*
    real 1m40,377s
    user 0m9,672s
    sys 0m19,928s

    *after*
    real 0m17,007s
    user 0m5,311s
    sys 0m11,779

    (5.8 times faster)

In the future, when all pernet_operations become async,
we'll just remove this "async" field tree-wide.

All the new logic is concentrated in patches [1-5/32].
The rest of patches converts specific operations:
review, rationale of they can be converted, and setting
of async flag.

Kirill

v3: Improved patches descriptions. Added comment into [5/32].
Added [32/32] converting netlink_tap_net_ops (new pernet operations
introduced in 2018).

v2: Single patch -> patchset with rationale of every conversion
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert netlink_tap_net_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:31:01 +0000 (12:31 +0300)]
net: Convert netlink_tap_net_ops

These pernet_operations init just allocated net memory,
and they obviously can be executed in parallel in any
others.

v3: New

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert diag_net_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:30:52 +0000 (12:30 +0300)]
net: Convert diag_net_ops

These pernet operations just create and destroy netlink
socket. The socket is pernet and else operations don't
touch it.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert default_device_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:30:42 +0000 (12:30 +0300)]
net: Convert default_device_ops

These pernet operations consist of exit() and exit_batch() methods.

default_device_exit() moves not-local and virtual devices to init_net.
There is nothing exciting, because this may happen in any time
on a working system, and rtnl_lock() and synchronize_net() protect
us from all cases of external dereference.

The same for default_device_exit_batch(). Similar unregisteration
may happen in any time on a system. Here several lists (like todo_list),
which are accessed under rtnl_lock(). After rtnl_unlock() and
netdev_run_todo() all the devices are flushed.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert loopback_net_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:30:35 +0000 (12:30 +0300)]
net: Convert loopback_net_ops

These pernet_operations have only init() method. It allocates
memory for net_device, calls register_netdev() and assigns
net::loopback_dev.

register_netdev() is allowed be used without additional locks,
as it's synchronized on rtnl_lock(). There are many examples
of using this functon directly from ioctl().

The only difference, compared to ioctl(), is that net is not
completely alive at this moment. But it looks like, there is
no way for parallel pernet_operations to dereference
the net_device, as the most of struct net_device lists,
where it's linked, are related to net, and the net is not liked.

The exceptions are net_device::unreg_list, close_list, todo_list,
used for unregistration, and ::link_watch_list, where net_device
may be linked to global lists.

Unregistration of loopback_dev obviously can't happen, when
loopback_net_init() is executing, as the net as alive. It occurs
in default_device_ops, which currently requires net_mutex,
and it behaves as a barrier at the moment. It will be considered
in next patch.

Speaking about link_watch_list, it seems, there is no way
for loopback_dev at time of registration to be linked in lweventlist
and be available for another pernet_operations.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert addrconf_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:30:27 +0000 (12:30 +0300)]
net: Convert addrconf_ops

These pernet_operations (un)register sysctl, which
are not touched by anybody else.

So, it's safe to make them async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert ipv4_sysctl_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:30:18 +0000 (12:30 +0300)]
net: Convert ipv4_sysctl_ops

These pernet_operations create and destroy sysctl,
which are not touched by anybody else.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert packet_net_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:30:08 +0000 (12:30 +0300)]
net: Convert packet_net_ops

These pernet_operations just create and destroy /proc entry,
and another operations do not touch it.

Also, nobody else are interested in foreign net::packet::sklist.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert unix_net_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:30:00 +0000 (12:30 +0300)]
net: Convert unix_net_ops

These pernet_operations are just create and destroy
/proc and sysctl entries, and are not touched by
foreign pernet_operations.

So, we are able to make them async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert pernet_subsys, registered from inet_init()
Kirill Tkhai [Tue, 13 Feb 2018 09:29:52 +0000 (12:29 +0300)]
net: Convert pernet_subsys, registered from inet_init()

arp_net_ops just addr/removes /proc entry.

devinet_ops allocates and frees duplicate of init_net tables
and (un)registers sysctl entries.

fib_net_ops allocates and frees pernet tables, creates/destroys
netlink socket and (un)initializes /proc entries. Foreign
pernet_operations do not touch them.

ip_rt_proc_ops only modifies pernet /proc entries.

xfrm_net_ops creates/destroys /proc entries, allocates/frees
pernet statistics, hashes and tables, and (un)initializes
sysctl files. These are not touched by foreigh pernet_operations

xfrm4_net_ops allocates/frees private pernet memory, and
configures sysctls.

sysctl_route_ops creates/destroys sysctls.

rt_genid_ops only initializes fields of just allocated net.

ipv4_inetpeer_ops allocated/frees net private memory.

igmp_net_ops just creates/destroys /proc files and socket,
noone else interested in.

tcp_sk_ops seems to be safe, because tcp_sk_init() does not
depend on any other pernet_operations modifications. Iteration
over hash table in inet_twsk_purge() is made under RCU lock,
and it's safe to iterate the table this way. Removing from
the table happen from inet_twsk_deschedule_put(), but this
function is safe without any extern locks, as it's synchronized
inside itself. There are many examples, it's used in different
context. So, it's safe to leave tcp_sk_exit_batch() unlocked.

tcp_net_metrics_ops is synchronized on tcp_metrics_lock and safe.

udplite4_net_ops only creates/destroys pernet /proc file.

icmp_sk_ops creates percpu sockets, not touched by foreign
pernet_operations.

ipmr_net_ops creates/destroys pernet fib tables, (un)registers
fib rules and /proc files. This seem to be safe to execute
in parallel with foreign pernet_operations.

af_inet_ops just sets up default parameters of newly created net.

ipv4_mib_ops creates and destroys pernet percpu statistics.

raw_net_ops, tcp4_net_ops, udp4_net_ops, ping_v4_net_ops
and ip_proc_ops only create/destroy pernet /proc files.

ip4_frags_ops creates and destroys sysctl file.

So, it's safe to make the pernet_operations async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert sysctl_core_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:29:42 +0000 (12:29 +0300)]
net: Convert sysctl_core_ops

These pernet_operations register and destroy sysctl
directory, and it's not interesting for foreign
pernet_operations.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert wext_pernet_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:29:33 +0000 (12:29 +0300)]
net: Convert wext_pernet_ops

These pernet_operations initialize and purge net::wext_nlevents
queue, and are not touched by foreign pernet_operations.

Mark them async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert genl_pernet_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:29:23 +0000 (12:29 +0300)]
net: Convert genl_pernet_ops

This pernet_operations create and destroy net::genl_sock.
Foreign pernet_operations don't touch it.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert subsys_initcall() registered pernet_operations from net/sched
Kirill Tkhai [Tue, 13 Feb 2018 09:29:13 +0000 (12:29 +0300)]
net: Convert subsys_initcall() registered pernet_operations from net/sched

psched_net_ops only creates and destroyes /proc entry,
and safe to be executed in parallel with any foreigh
pernet_operations.

tcf_action_net_ops initializes and destructs tcf_action_net::egdev_ht,
which is not touched by foreign pernet_operations.

So, make them async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert fib_* pernet_operations, registered via subsys_initcall
Kirill Tkhai [Tue, 13 Feb 2018 09:29:03 +0000 (12:29 +0300)]
net: Convert fib_* pernet_operations, registered via subsys_initcall

Both of them create and initialize lists, which are not touched
by another foreing pernet_operations.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert pernet_subsys ops, registered via net_dev_init()
Kirill Tkhai [Tue, 13 Feb 2018 09:28:54 +0000 (12:28 +0300)]
net: Convert pernet_subsys ops, registered via net_dev_init()

There are:
1)dev_proc_ops and dev_mc_net_ops, which create and destroy
pernet proc file and not interesting for another net namespaces;
2)netdev_net_ops, which creates pernet hashes, which are not
touched by another pernet_operations.

So, make them async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert proto_net_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:28:44 +0000 (12:28 +0300)]
net: Convert proto_net_ops

This patch starts to convert pernet_subsys, registered
from subsys initcalls.

It seems safe to be executed in parallel with others,
as it's only creates/destoyes proc entry,
which nobody else is not interested in.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert uevent_net_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:28:33 +0000 (12:28 +0300)]
net: Convert uevent_net_ops

uevent_net_init() and uevent_net_exit() create and
destroy netlink socket, and these actions serialized
in netlink code.

Parallel execution with other pernet_operations
makes the socket disappear earlier from uevent_sock_list
on ->exit. As userspace can't be interested in broadcast
messages of dying net, and, as I see, no one in kernel
listen them, we may safely make uevent_net_ops async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert audit_net_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:28:24 +0000 (12:28 +0300)]
net: Convert audit_net_ops

This patch starts to convert pernet_subsys, registered
from postcore initcalls.

audit_net_init() creates netlink socket, while audit_net_exit()
destroys it. The rest of the pernet_list are not interested
in the socket, so we make audit_net_ops async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert rtnetlink_net_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:28:15 +0000 (12:28 +0300)]
net: Convert rtnetlink_net_ops

rtnetlink_net_init() and rtnetlink_net_exit()
create and destroy netlink socket net::rtnl.

The socket is used to send rtnl notification via
rtnl_net_notifyid(). There is no a problem
to create and destroy it in parallel with other
pernet operations, as we link net in setup_net()
after the socket is created, and destroy
in cleanup_net() after net is unhashed from all
the lists and there is no RCU references on it.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert netlink_net_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:28:00 +0000 (12:28 +0300)]
net: Convert netlink_net_ops

The methods of netlink_net_ops create and destroy "netlink"
file, which are not interesting for foreigh pernet_operations.
So, netlink_net_ops may safely be made async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert net_defaults_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:27:51 +0000 (12:27 +0300)]
net: Convert net_defaults_ops

net_defaults_ops introduce only net_defaults_init_net method,
and it acts on net::core::sysctl_somaxconn, which
is not interesting for the rest of pernet_subsys and
pernet_device lists. Then, make them async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert net_inuse_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:27:41 +0000 (12:27 +0300)]
net: Convert net_inuse_ops

net_inuse_ops methods expose statistics in /proc.
No one from the rest of pernet_subsys or pernet_device
lists touch net::core::inuse.

So, it's safe to make net_inuse_ops async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert nf_log_net_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:27:31 +0000 (12:27 +0300)]
net: Convert nf_log_net_ops

The pernet_operations would have had a problem in parallel
execution with others, if init_net had been able to released.
But it's not, and the rest is safe for that.
There is memory allocation, which nobody else interested in,
and sysctl registration. So, we make them async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert netfilter_net_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:27:23 +0000 (12:27 +0300)]
net: Convert netfilter_net_ops

Methods netfilter_net_init() and netfilter_net_exit()
initialize net::nf::hooks and change net-related proc
directory of net. Another pernet_operations are not
interested in forein net::nf::hooks or proc entries,
so it's safe to make them executed in parallel with
methods of other pernet operations.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert sysctl_pernet_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:27:13 +0000 (12:27 +0300)]
net: Convert sysctl_pernet_ops

This patch starts to convert pernet_subsys, registered
from core initcalls.

Methods sysctl_net_init() and sysctl_net_exit() initialize
net::sysctls table of a namespace.

pernet_operations::init()/exit() methods from the rest
of the list do not touch net::sysctls of strangers,
so it's safe to execute sysctl_pernet_ops's methods
in parallel with any other pernet_operations.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert net_ns_ops methods
Kirill Tkhai [Tue, 13 Feb 2018 09:27:03 +0000 (12:27 +0300)]
net: Convert net_ns_ops methods

This patch starts to convert pernet_subsys, registered
from pure initcalls.

net_ns_ops::net_ns_net_init/net_ns_net_init, methods use only
ida_simple_* functions, which are not need a synchronization.
They are synchronized by idr subsystem.

So, net_ns_ops methods are able to be executed
in parallel with methods of other pernet operations.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert proc_net_ns_ops
Kirill Tkhai [Tue, 13 Feb 2018 09:26:52 +0000 (12:26 +0300)]
net: Convert proc_net_ns_ops

This patch starts to convert pernet_subsys, registered
before initcalls.

proc_net_ns_ops::proc_net_ns_init()/proc_net_ns_exit()
{un,}register pernet net->proc_net and ->proc_net_stat.

Constructors and destructors of another pernet_operations
are not interested in foreign net's proc_net and proc_net_stat.
Proc filesystem privitives are synchronized on proc_subdir_lock.

So, proc_net_ns_ops methods are able to be executed
in parallel with methods of any other pernet operations.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Allow pernet_operations to be executed in parallel
Kirill Tkhai [Tue, 13 Feb 2018 09:26:44 +0000 (12:26 +0300)]
net: Allow pernet_operations to be executed in parallel

This adds new pernet_operations::async flag to indicate operations,
which ->init(), ->exit() and ->exit_batch() methods are allowed
to be executed in parallel with the methods of any other pernet_operations.

When there are only asynchronous pernet_operations in the system,
net_mutex won't be taken for a net construction and destruction.

Also, remove BUG_ON(mutex_is_locked()) from net_assign_generic()
without replacing with the equivalent net_sem check, as there is
one more lockdep assert below.

v3: Add comment near net_mutex.

Suggested-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>