openwrt/staging/blogic.git
5 years agonet: ethernet: emac: Fix phy mode type
Andrew Lunn [Tue, 5 Nov 2019 17:53:23 +0000 (18:53 +0100)]
net: ethernet: emac: Fix phy mode type

Pass a phy_interface_t to of_get_phy_mode(), by changing the type of
phy_mode in the device structure. This then requires that
zmii_attach() is also changes, since it takes a pointer to phy_mode.

Fixes: 0c65b2b90d13 ("net: of_get_phy_mode: Change API to solve int/unit warnings")
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net_sched-convert-packet-counters-to-64bit'
David S. Miller [Wed, 6 Nov 2019 02:20:55 +0000 (18:20 -0800)]
Merge branch 'net_sched-convert-packet-counters-to-64bit'

Eric Dumazet says:

====================
net_sched: convert packet counters to 64bit

This small patch series add 64bit support for packet counts.

Fact that the counters were still 32bit has been quite painful.

tc -s -d qd sh dev eth0 | head -3
qdisc mq 1: root
 Sent 665706335338 bytes 6526520373 pkt (dropped 2441, overlimits 0 requeues 91)
 backlog 0b 0p requeues 91
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet_sched: add TCA_STATS_PKT64 attribute
Eric Dumazet [Tue, 5 Nov 2019 03:13:15 +0000 (19:13 -0800)]
net_sched: add TCA_STATS_PKT64 attribute

Now the kernel uses 64bit packet counters in scheduler layer,
we want to export these counters to user space.

Instead risking breaking user space by adding fields
to struct gnet_stats_basic, add a new TCA_STATS_PKT64.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet_sched: extend packet counter to 64bit
Eric Dumazet [Tue, 5 Nov 2019 03:13:14 +0000 (19:13 -0800)]
net_sched: extend packet counter to 64bit

After this change, qdisc packet counter is no longer
a 32bit quantity. We still export 32bit values to user.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet_sched: do not export gnet_stats_basic_packed to uapi
Eric Dumazet [Tue, 5 Nov 2019 03:13:13 +0000 (19:13 -0800)]
net_sched: do not export gnet_stats_basic_packed to uapi

gnet_stats_basic_packed was really meant to be private kernel structure.

If this proves to be a problem, we will have to rename the in-kernel
version.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'mv88e6xxx-ATU-occupancy-as-devlink-resource'
David S. Miller [Wed, 6 Nov 2019 02:09:45 +0000 (18:09 -0800)]
Merge branch 'mv88e6xxx-ATU-occupancy-as-devlink-resource'

Andrew Lunn says:

====================
mv88e6xxx ATU occupancy as devlink resource

This patchset add generic support to DSA for devlink resources. The
Marvell switch Address Translation Unit occupancy is then exported as
a resource. In order to do this, the number of ATU entries is added to
the per switch info structure. Helpers are added, and then the
resource itself is then added.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: Add ATU occupancy via devlink resources
Andrew Lunn [Tue, 5 Nov 2019 00:13:01 +0000 (01:13 +0100)]
net: dsa: mv88e6xxx: Add ATU occupancy via devlink resources

The ATU can report how many entries it contains. It does this per bin,
there being 4 bins in total. Export the ATU as a devlink resource, and
provide a method the needed callback to get the resource occupancy.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: global1_atu: Add helper for get next
Andrew Lunn [Tue, 5 Nov 2019 00:13:00 +0000 (01:13 +0100)]
net: dsa: mv88e6xxx: global1_atu: Add helper for get next

When retrieving the ATU statistics, and ATU get next has to be
performed to trigger the ATU to collect the statistics. Export a
helper from global1_atu to perform this.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: global2: Expose ATU stats register
Andrew Lunn [Tue, 5 Nov 2019 00:12:59 +0000 (01:12 +0100)]
net: dsa: mv88e6xxx: global2: Expose ATU stats register

Add helpers to set/get the ATU statistics register.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: Add number of MACs in the ATU
Andrew Lunn [Tue, 5 Nov 2019 00:12:58 +0000 (01:12 +0100)]
net: dsa: mv88e6xxx: Add number of MACs in the ATU

For each supported switch, add an entry to the info structure for the
number of MACs which can be stored in the ATU. This will later be used
to export the ATU as a devlink resource, and indicate its occupancy,
how full the ATU is.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: Add support for devlink resources
Andrew Lunn [Tue, 5 Nov 2019 00:12:57 +0000 (01:12 +0100)]
net: dsa: Add support for devlink resources

Add wrappers around the devlink resource API, so that DSA drivers can
register and unregister devlink resources.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-dsa-bcm_sf2-Add-support-for-optional-reset-controller-line'
David S. Miller [Wed, 6 Nov 2019 02:06:38 +0000 (18:06 -0800)]
Merge branch 'net-dsa-bcm_sf2-Add-support-for-optional-reset-controller-line'

Florian Fainelli says:

====================
net: dsa: bcm_sf2: Add support for optional reset controller line

This patch series definest the optional reset controller line for the
BCM7445/BCM7278 integrated Ethernet switches and updates the driver to
drive that reset line in lieu of the internal watchdog based reset since
it does not work on BCM7278.

Changes in v2:
- make the reset_control_assert() conditional to BCM7278 in the remove
  function as well
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: bcm_sf2: Add support for optional reset controller line
Florian Fainelli [Mon, 4 Nov 2019 21:51:39 +0000 (13:51 -0800)]
net: dsa: bcm_sf2: Add support for optional reset controller line

Grab an optional and exclusive reset controller line for the switch and
manage it during probe/remove functions accordingly. For 7278 devices we
change bcm_sf2_sw_rst() to use the reset controller line since the
WATCHDOG_CTRL register does not reset the switch contrary to stated
documentation.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodt-bindings: net: Describe BCM7445 switch reset property
Florian Fainelli [Mon, 4 Nov 2019 21:51:38 +0000 (13:51 -0800)]
dt-bindings: net: Describe BCM7445 switch reset property

The BCM7445/BCM7278 built-in Ethernet switch have an optional reset line
to the SoC's reset controller, describe the 'resets' and 'reset-names'
properties as optional.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoChange in Openvswitch to support MPLS label depth of 3 in ingress direction
Martin Varghese [Mon, 4 Nov 2019 01:57:44 +0000 (07:27 +0530)]
Change in Openvswitch to support MPLS label depth of 3 in ingress direction

The openvswitch was supporting a MPLS label depth of 1 in the ingress
direction though the userspace OVS supports a max depth of 3 labels.
This change enables openvswitch module to support a max depth of
3 labels in the ingress.

Signed-off-by: Martin Varghese <martin.varghese@nokia.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: remove unused macros
Colin Ian King [Sun, 3 Nov 2019 13:15:38 +0000 (13:15 +0000)]
net: hns3: remove unused macros

The macros HCLGE_MPF_ENBALE and HCLGEVF_MPF_ENBALE are defined but never
used.  I was going to fix the spelling mistake "ENBALE" -> "ENABLE" but
found these macros are not used, so they can be removed.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agovsock: Simplify '__vsock_release()'
Christophe JAILLET [Sun, 3 Nov 2019 06:11:11 +0000 (07:11 +0100)]
vsock: Simplify '__vsock_release()'

Use 'skb_queue_purge()' instead of re-implementing it.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: Fix use after free in dsa_switch_remove()
Florian Fainelli [Sun, 3 Nov 2019 03:13:26 +0000 (20:13 -0700)]
net: dsa: Fix use after free in dsa_switch_remove()

The order in which the ports are deleted from the list and freed and the
call to dsa_switch_remove() is done is reversed, which leads to an
use after free condition. Reverse the two: first tear down the ports and
switch from the fabric, then free the ports associated with that switch
fabric.

Fixes: 05f294a85235 ("net: dsa: allocate ports on touch")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotc-testing: added tests with cookie for mpls TC action
Roman Mashak [Sat, 2 Nov 2019 22:25:51 +0000 (18:25 -0400)]
tc-testing: added tests with cookie for mpls TC action

Signed-off-by: Roman Mashak <mrv@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'icmp-move-duplicate-code-in-helper-functions'
David S. Miller [Tue, 5 Nov 2019 22:03:11 +0000 (14:03 -0800)]
Merge branch 'icmp-move-duplicate-code-in-helper-functions'

Matteo Croce says:

====================
icmp: move duplicate code in helper functions

Remove some duplicate code by moving it in two helper functions.
First patch adds the helpers, the second one uses it.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoicmp: remove duplicate code
Matteo Croce [Sat, 2 Nov 2019 00:12:04 +0000 (01:12 +0100)]
icmp: remove duplicate code

The same code which recognizes ICMP error packets is duplicated several
times. Use the icmp_is_err() and icmpv6_is_err() helpers instead, which
do the same thing.

ip_multipath_l3_keys() and tcf_nat_act() didn't check for all the error types,
assume that they should instead.

Signed-off-by: Matteo Croce <mcroce@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoicmp: add helpers to recognize ICMP error packets
Matteo Croce [Sat, 2 Nov 2019 00:12:03 +0000 (01:12 +0100)]
icmp: add helpers to recognize ICMP error packets

Add two helper functions, one for IPv4 and one for IPv6, to recognize
the ICMP packets which are error responses.
This packets are special because they have as payload the original
header of the packet which generated it (RFC 792 says at least 8 bytes,
but Linux actually includes much more than that).

Signed-off-by: Matteo Croce <mcroce@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'netvsc-RSS-related-patches'
David S. Miller [Tue, 5 Nov 2019 22:01:35 +0000 (14:01 -0800)]
Merge branch 'netvsc-RSS-related-patches'

Stephen Hemminger says:

====================
netvsc: RSS related patches

Address a couple of issues related to recording RSS hash
value in skb. These were found by reviewing RSS support.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agohv_netvsc: record hardware hash in skb
Stephen Hemminger [Fri, 1 Nov 2019 23:42:38 +0000 (16:42 -0700)]
hv_netvsc: record hardware hash in skb

Since RSS hash is available from the host, record it in
the skb.

Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agohv_netvsc: flag software created hash value
Stephen Hemminger [Fri, 1 Nov 2019 23:42:37 +0000 (16:42 -0700)]
hv_netvsc: flag software created hash value

When the driver needs to create a hash value because it
was not done at higher level, then the hash should be marked
as a software not hardware hash.

Fixes: f72860afa2e3 ("hv_netvsc: Exclude non-TCP port numbers from vRSS hashing")
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Tue, 5 Nov 2019 21:40:12 +0000 (13:40 -0800)]
Merge branch '100GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
100GbE Intel Wired LAN Driver Updates 2019-11-04

This series contains updates to the ice driver only.

Anirudh refactors the code to reduce the kernel configuration flags and
introduces ice_base.c file.

Maciej does additional refactoring on the configuring of transmit
rings so that we are not configuring per each traffic class flow.
Added support for XDP in the ice driver.  Provides additional
re-organizing of the code in preparation for adding build_skb() support
in the driver.  Adjusted the computational padding logic for headroom
and tailroom to better support build_skb(), which also aligns with the
logic in other Intel LAN drivers.  Added build_skb support and make use
of the XDP's data_meta.

Krzysztof refactors the driver to prepare for AF_XDP support in the
driver and then adds support for AF_XDP.

v2: Updated patch 3 of the series based on community feedback with the
    following changes...
    - return -EOPNOTSUPP instead of ENOTSUPP for too large MTU which makes
      it impossible to attach XDP prog
    - don't check for case when there's no XDP prog currently on interface
      and ice_xdp() is called with NULL bpf_prog; this happens when user
      does "ip link set eth0 xdp off" and no prog is present on VSI; no need
      for that as it is handled by higher layer
    - drop the extack message for unknown xdp->command
    - use the smp_processor_id() for accessing the XDP Tx ring for XDP_TX
      action
    - don't leave the interface in downed state in case of any failure
      during the XDP Tx resources handling
    - undo rename of ice_build_ctob
    The above changes caused a ripple effect in patches 4 & 5 to update
    references to ice_build_ctob() which are now build_ctob()
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch '10GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Tue, 5 Nov 2019 21:36:31 +0000 (13:36 -0800)]
Merge branch '10GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
10GbE Intel Wired LAN Driver Updates 2019-11-04

This series contains old Halloween candy updates, yet still sweet, to
fm10k, ixgbe and i40e.

Jake adds the missing initializers for a couple of the TLV attribute
macros.  Added support for capturing and reporting statistics for all of
the VFs in a given PF.  Lastly, bump the version of the fm10k driver to
reflect the recent changes.

Alex addresses locality issues in the ixgbe driver when it is loaded on
a system supporting multiple NUMA nodes.

Manjunath Patil provides changes to the ixgbe driver, similar to those
made to igb, to prevent transmit packets to request a hardware timestamp
when the NIC has not been setup via the SIOCSHWTSTAMP ioctl.

Alice adds support for x710 by adding the missing device id's in the
appropriate places to ensure all the features are enabled in i40e.

Jesse adds support for VF stats gathering in the i40e via the kernel
via ndo_get_vf_stats function.

v2: Fixed up commit id references in patch 5's description to align with
    how commit id's should be referenced.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoi40e: implement VF stats NDO
Jesse Brandeburg [Tue, 29 Oct 2019 00:37:07 +0000 (17:37 -0700)]
i40e: implement VF stats NDO

Implement the VF stats gathering via the kernel via ndo_get_vf_stats().
The driver will show per-VF stats in the output of the command:
ip -s link show dev <PF>

Testing Hints:
ip -s link show dev eth0
will return non-zero VF stats.
...
   vf 0 MAC 00:55:aa:00:55:aa, spoof checking on, link-state enable, trust off
   RX: bytes  packets  mcast   bcast
   128000     1000     104     104
   TX: bytes  packets
   128000     1000

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoi40e: enable X710 support
Alice Michael [Wed, 23 Oct 2019 10:19:57 +0000 (03:19 -0700)]
i40e: enable X710 support

The I40E_DEV_ID_10G_BASE_T_BC device id was added previously,
but was not enabled in all the appropriate places.  Adding it
to enable it's use.

Signed-off-by: Alice Michael <alice.michael@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoixgbe: protect TX timestamping from API misuse
Manjunath Patil [Sat, 5 Oct 2019 15:20:03 +0000 (08:20 -0700)]
ixgbe: protect TX timestamping from API misuse

HW timestamping can only be requested for a packet if the NIC is first
setup via ioctl(SIOCSHWTSTAMP). If this step was skipped, then the ixgbe
driver still allowed TX packets to request HW timestamping. In this
situation, we see 'clearing Tx Timestamp hang' noise in the log.

Fix this by checking that the NIC is configured for HW TX timestamping
before accepting a HW TX timestamping request.

Similar-to:
   commit 26bd4e2db06b ("igb: protect TX timestamping from API misuse")
   commit 0a6f2f05a2f5 ("igb: Fix a test with HWTSTAMP_TX_ON")

Signed-off-by: Manjunath Patil <manjunath.b.patil@oracle.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agofm10k: update driver version to match out-of-tree
Jacob Keller [Thu, 26 Sep 2019 21:29:04 +0000 (14:29 -0700)]
fm10k: update driver version to match out-of-tree

An upcoming out-of-tree release will be occurring which will include the
recent functionality to support virtual function statistics. Update the
kernel driver version to match this.

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>
5 years agoixgbe: Make use of cpumask_local_spread to improve RSS locality
Alexander Duyck [Sat, 21 Sep 2019 00:18:50 +0000 (17:18 -0700)]
ixgbe: Make use of cpumask_local_spread to improve RSS locality

This patch is meant to address locality issues present in the ixgbe driver
when it is loaded on a system supporting multiple NUMA nodes and more CPUs
then the device can map in a 1:1 fashion. Instead of just arbitrarily
mapping itself to CPUs 0-62 it would make much more sense to map itself to
the local CPUs first, and then map itself to any remaining CPUs that might
be used.

The first effect of this is that queue 0 should always be allocated on the
local CPU/NUMA node. This is important as it is the default destination if
a packet doesn't match any existing flow director filter or RSS rule and as
such having it local should help to reduce QPI cross-talk in the event of
an unrecognized traffic type.

In addition this should increase the likelihood of the RSS queues being
allocated and used on CPUs local to the device while the ATR/Flow Director
queues would be able to route traffic directly to the CPU that is likely to
be processing it.

Signed-off-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agofm10k: add support for ndo_get_vf_stats operation
Jacob Keller [Tue, 20 Aug 2019 21:19:21 +0000 (14:19 -0700)]
fm10k: add support for ndo_get_vf_stats operation

Support capturing and reporting statistics for all of the VFs associated
with a given PF device via the ndo_get_vf_stats callback.

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>
5 years agofm10k: add missing field initializers to TLV attributes)
Jacob Keller [Tue, 19 Feb 2019 22:26:27 +0000 (14:26 -0800)]
fm10k: add missing field initializers to TLV attributes)

Add the missing field initializers for a couple of the TLV attribute
macros. This resolves the last few -Wmissing-field-initializers warnings
for the fm10k Linux driver.

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>
5 years agoice: allow 3k MTU for XDP
Maciej Fijalkowski [Thu, 24 Oct 2019 08:11:25 +0000 (01:11 -0700)]
ice: allow 3k MTU for XDP

At this point ice driver is able to work on order 1 pages that are split
onto two 3k buffers. Let's reflect that when user is setting new MTU
size and XDP is present on interface.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: add build_skb() support
Maciej Fijalkowski [Thu, 24 Oct 2019 08:11:24 +0000 (01:11 -0700)]
ice: add build_skb() support

Driver is now prepared for building the skb around the existing Rx
buffer, so introduce the ice_build_skb responsible for it. Make use of
XDP's data_meta as well.

I've observed around 30% less CPU consumption with build_skb Rx path, in
comparison to legacy Rx. What stands behind such result is the avoidance
of flow_dissector (which we were diving into via eth_get_headlen) and no
memcpy calls.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: introduce frame padding computation logic
Maciej Fijalkowski [Thu, 24 Oct 2019 08:11:23 +0000 (01:11 -0700)]
ice: introduce frame padding computation logic

Take into account the underlying architecture specific settings and
based on that calculate the possible padding that can be supplied.
Typically, for x86 and standard MTU size we will end up with 192 bytes
of headroom. This is the same behavior as our other drivers have and we
can dedicate it for XDP purposes.

Furthermore, introduce the Rx ring flag for indicating whether build_skb
is used on particular. Based on that invoke the routines for padding
calculation.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: introduce legacy Rx flag
Maciej Fijalkowski [Thu, 24 Oct 2019 08:11:22 +0000 (01:11 -0700)]
ice: introduce legacy Rx flag

Add an ethtool "legacy-rx" priv flag for toggling the Rx path. This
control knob will be mainly used for build_skb usage as well as buffer
size/MTU manipulation.

In preparation for adding build_skb support in a way that it takes
care of how we set the values of max_frame and rx_buf_len fields of
struct ice_vsi. Specifically, in this patch mentioned fields are set to
values that will allow us to provide headroom and tailroom in-place.

This can be mostly broken down onto following:
- for legacy-rx "on" ethtool control knob, old behaviour is kept;
- for standard 1500 MTU size configure the buffer of size 1536, as
  network stack is expecting the NET_SKB_PAD to be provided and
  NET_IP_ALIGN can have a non-zero value (these can be typically equal
  to 32 and 2, respectively);
- for larger MTUs go with max_frame set to 9k and configure the 3k
  buffer in case when PAGE_SIZE of underlying arch is less than 8k; 3k
  buffer is implying the need for order 1 page, so that our page
  recycling scheme can still be applied;

With that said, substitute the hardcoded ICE_RXBUF_2048 and PAGE_SIZE
values in DMA API that we're making use of with rx_ring->rx_buf_len and
ice_rx_pg_size(rx_ring). The latter is an introduced helper for
determining the page size based on its order (which was figured out via
ice_rx_pg_order). Last but not least, take care of truesize calculation.

In the followup patch the headroom/tailroom computation logic will be
introduced.

This change aligns the buffer and frame configuration with other Intel
drivers, most importantly with iavf.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Add support for AF_XDP
Krzysztof Kazimierczak [Mon, 4 Nov 2019 17:38:56 +0000 (09:38 -0800)]
ice: Add support for AF_XDP

Add zero copy AF_XDP support.  This patch adds zero copy support for
Tx and Rx; code for zero copy is added to ice_xsk.h and ice_xsk.c.

For Tx, implement ndo_xsk_wakeup. As with other drivers, reuse
existing XDP Tx queues for this task, since XDP_REDIRECT guarantees
mutual exclusion between different NAPI contexts based on CPU ID. In
turn, a netdev can XDP_REDIRECT to another netdev with a different
NAPI context, since the operation is bound to a specific core and each
core has its own hardware ring.

For Rx, allocate frames as MEM_TYPE_ZERO_COPY on queues that AF_XDP is
enabled.

Signed-off-by: Krzysztof Kazimierczak <krzysztof.kazimierczak@intel.com>
Co-developed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Move common functions to ice_txrx_lib.c
Krzysztof Kazimierczak [Mon, 4 Nov 2019 17:38:56 +0000 (09:38 -0800)]
ice: Move common functions to ice_txrx_lib.c

In preparation of AF XDP, move functions that will be used both by skb and
zero-copy paths to a new file called ice_txrx_lib.c.  This allows us to
avoid using ifdefs to control the staticness of said functions.

Move other functions (ice_rx_csum, ice_rx_hash and ice_ptype_to_htype)
called only by the moved ones to the new file as well.

Signed-off-by: Krzysztof Kazimierczak <krzysztof.kazimierczak@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agobpf: re-fix skip write only files in debugfs
Daniel Borkmann [Mon, 4 Nov 2019 14:27:02 +0000 (15:27 +0100)]
bpf: re-fix skip write only files in debugfs

Commit 5bc60de50dfe ("selftests: bpf: Don't try to read files without
read permission") got reverted as the fix was not working as expected
and real fix came in via 8101e069418d ("selftests: bpf: Skip write
only files in debugfs"). When bpf-next got merged into net-next, the
test_offload.py had a small conflict. Fix the resolution in ae8a76fb8b5d
iby not reintroducing 5bc60de50dfe again.

Fixes: ae8a76fb8b5d ("Merge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jakub Kicinski <jakub.kicinski@netronome.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: stmmac: drop unused variable in stm32mp1_set_mode()
Christophe Roullier [Mon, 4 Nov 2019 10:51:00 +0000 (11:51 +0100)]
net: ethernet: stmmac: drop unused variable in stm32mp1_set_mode()

Building with W=1 (cf.scripts/Makefile.extrawarn) outputs:
warning: variable ‘ret’ set but not used [-Wunused-but-set-variable]

Drop the unused 'ret' variable.

Signed-off-by: Christophe Roullier <christophe.roullier@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: sgi: ioc3-eth: ensure tx ring is 16k aligned.
Thomas Bogendoerfer [Mon, 4 Nov 2019 10:45:15 +0000 (11:45 +0100)]
net: sgi: ioc3-eth: ensure tx ring is 16k aligned.

IOC3 hardware needs a 16k aligned TX ring.

Signed-off-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: sgi: ioc3-eth: fix setting NETIF_F_HIGHDMA
Christoph Hellwig [Mon, 4 Nov 2019 10:45:14 +0000 (11:45 +0100)]
net: sgi: ioc3-eth: fix setting NETIF_F_HIGHDMA

Set NETIF_F_HIGHDMA together with the NETIF_F_IP_CSUM flag instead of
letting the second assignment overwrite it.  Probably doesn't matter
in practice as none of the systems an IOC3 is usually found in has
highmem to start with.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: sgi: ioc3-eth: simplify setting the DMA mask
Christoph Hellwig [Mon, 4 Nov 2019 10:45:13 +0000 (11:45 +0100)]
net: sgi: ioc3-eth: simplify setting the DMA mask

There is no need to fall back to a lower mask these days, the DMA mask
just communicates the hardware supported features.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: sgi: ioc3-eth: fix usage of GFP_* flags
Christoph Hellwig [Mon, 4 Nov 2019 10:45:12 +0000 (11:45 +0100)]
net: sgi: ioc3-eth: fix usage of GFP_* flags

dma_alloc_coherent always zeroes memory, there is no need for
__GFP_ZERO.  Also doing a GFP_ATOMIC allocation just before a GFP_KERNEL
one is clearly bogus.

Fixes: ed870f6a7aa2 ("net: sgi: ioc3-eth: use dma-direct for dma allocations")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: sgi: ioc3-eth: don't abuse dma_direct_* calls
Christoph Hellwig [Mon, 4 Nov 2019 10:45:11 +0000 (11:45 +0100)]
net: sgi: ioc3-eth: don't abuse dma_direct_* calls

dma_direct_ is a low-level API that must never be used by drivers
directly.  Switch to use the proper DMA API instead.

Fixes: ed870f6a7aa2 ("net: sgi: ioc3-eth: use dma-direct for dma allocations")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: use jhash2() in rt6_exception_hash()
Eric Dumazet [Mon, 4 Nov 2019 02:24:16 +0000 (18:24 -0800)]
ipv6: use jhash2() in rt6_exception_hash()

Faster jhash2() can be used instead of jhash(), since
IPv6 addresses have the needed alignment requirement.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: of_get_phy_mode: Change API to solve int/unit warnings
Andrew Lunn [Mon, 4 Nov 2019 01:40:33 +0000 (02:40 +0100)]
net: of_get_phy_mode: Change API to solve int/unit warnings

Before this change of_get_phy_mode() returned an enum,
phy_interface_t. On error, -ENODEV etc, is returned. If the result of
the function is stored in a variable of type phy_interface_t, and the
compiler has decided to represent this as an unsigned int, comparision
with -ENODEV etc, is a signed vs unsigned comparision.

Fix this problem by changing the API. Make the function return an
error, or 0 on success, and pass a pointer, of type phy_interface_t,
where the phy mode should be stored.

v2:
Return with *interface set to PHY_INTERFACE_MODE_NA on error.
Add error checks to all users of of_get_phy_mode()
Fixup a few reverse christmas tree errors
Fixup a few slightly malformed reverse christmas trees

v3:
Fix 0-day reported errors.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: bridge: fdb: eliminate extra port state tests from fast-path
Nikolay Aleksandrov [Mon, 4 Nov 2019 09:36:51 +0000 (11:36 +0200)]
net: bridge: fdb: eliminate extra port state tests from fast-path

When commit df1c0b8468b3 ("[BRIDGE]: Packets leaking out of
disabled/blocked ports.") introduced the port state tests in
br_fdb_update() it was to avoid learning/refreshing from STP BPDUs, it was
also used to avoid learning/refreshing from user-space with NTF_USE. Those
two tests are done for every packet entering the bridge if it's learning,
but for the fast-path we already have them checked in br_handle_frame() and
is unnecessary to do it again. Thus push the checks to the unlikely cases
and drop them from br_fdb_update(), the new nbp_state_should_learn() helper
is used to determine if the port state allows br_fdb_update() to be called.
The two places which need to do it manually are:
 - user-space add call with NTF_USE set
 - link-local packet learning done in __br_handle_local_finish()

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoice: Add support for XDP
Maciej Fijalkowski [Mon, 4 Nov 2019 17:38:56 +0000 (09:38 -0800)]
ice: Add support for XDP

Add support for XDP. Implement ndo_bpf and ndo_xdp_xmit.  Upon load of
an XDP program, allocate additional Tx rings for dedicated XDP use.
The following actions are supported: XDP_TX, XDP_DROP, XDP_REDIRECT,
XDP_PASS, and XDP_ABORTED.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: get rid of per-tc flow in Tx queue configuration routines
Maciej Fijalkowski [Thu, 24 Oct 2019 08:11:18 +0000 (01:11 -0700)]
ice: get rid of per-tc flow in Tx queue configuration routines

There's no reason for treating DCB as first class citizen when configuring
the Tx queues and going through TCs. Reverse the logic and base the
configuration logic on rings, which is the object of interest anyway.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Introduce ice_base.c
Anirudh Venkataramanan [Thu, 24 Oct 2019 08:11:17 +0000 (01:11 -0700)]
ice: Introduce ice_base.c

Remove a few uses of kernel configuration flags from ice_lib.c by
introducing a new source file ice_base.c. Also move corresponding
function prototypes from ice_lib.h to ice_base.h and include ice_base.h
where required.

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoMerge tag 'mlx5-updates-2019-11-01' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Mon, 4 Nov 2019 03:23:49 +0000 (19:23 -0800)]
Merge tag 'mlx5-updates-2019-11-01' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2019-11-01

Misc updates for mlx5 netdev and core driver

1) Steering Core: Replace CRC32 internal implementation with standard
   kernel lib.
2) Steering Core: Support IPv4 and IPv6 mixed matcher.
3) Steering Core: Lockless FTE read lookups
4) TC: Bit sized fields rewrite support.
5) FPGA: Standalone FPGA support.
6) SRIOV: Reset VF parameters configurations on SRIOV disable.
7) netdev: Dump WQs wqe descriptors on CQE with error events.
8) MISC Cleanups.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomISDN: remove unused variable 'faxmodulation_s'
YueHaibing [Fri, 1 Nov 2019 13:44:47 +0000 (21:44 +0800)]
mISDN: remove unused variable 'faxmodulation_s'

drivers/isdn/hardware/mISDN/mISDNisar.c:30:17:
 warning: faxmodulation_s defined but not used [-Wunused-const-variable=]

It is never used, so can be removed.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoptp: Add a ptp clock driver for IDT ClockMatrix.
Vincent Cheng [Fri, 1 Nov 2019 03:20:07 +0000 (23:20 -0400)]
ptp: Add a ptp clock driver for IDT ClockMatrix.

The IDT ClockMatrix (TM) family includes integrated devices that provide
eight PLL channels.  Each PLL channel can be independently configured as a
frequency synthesizer, jitter attenuator, digitally controlled
oscillator (DCO), or a digital phase lock loop (DPLL).  Typically
these devices are used as timing references and clock sources for PTP
applications.  This patch adds support for the device.

Co-developed-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Vincent Cheng <vincent.cheng.xh@renesas.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodt-bindings: ptp: Add device tree binding for IDT ClockMatrix based PTP clock
Vincent Cheng [Fri, 1 Nov 2019 03:20:06 +0000 (23:20 -0400)]
dt-bindings: ptp: Add device tree binding for IDT ClockMatrix based PTP clock

Add device tree binding doc for the IDT ClockMatrix PTP clock.

Signed-off-by: Vincent Cheng <vincent.cheng.xh@renesas.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: icmp6: provide input address for traceroute6
Francesco Ruggeri [Thu, 31 Oct 2019 00:40:02 +0000 (17:40 -0700)]
net: icmp6: provide input address for traceroute6

traceroute6 output can be confusing, in that it shows the address
that a router would use to reach the sender, rather than the address
the packet used to reach the router.
Consider this case:

        ------------------------ N2
         |                    |
       ------              ------  N3  ----
       | R1 |              | R2 |------|H2|
       ------              ------      ----
         |                    |
        ------------------------ N1
                  |
                 ----
                 |H1|
                 ----

where H1's default route is through R1, and R1's default route is
through R2 over N2.
traceroute6 from H1 to H2 shows R2's address on N1 rather than on N2.

The script below can be used to reproduce this scenario.

traceroute6 output without this patch:

traceroute to 2000:103::4 (2000:103::4), 30 hops max, 80 byte packets
 1  2000:101::1 (2000:101::1)  0.036 ms  0.008 ms  0.006 ms
 2  2000:101::2 (2000:101::2)  0.011 ms  0.008 ms  0.007 ms
 3  2000:103::4 (2000:103::4)  0.013 ms  0.010 ms  0.009 ms

traceroute6 output with this patch:

traceroute to 2000:103::4 (2000:103::4), 30 hops max, 80 byte packets
 1  2000:101::1 (2000:101::1)  0.056 ms  0.019 ms  0.006 ms
 2  2000:102::2 (2000:102::2)  0.013 ms  0.008 ms  0.008 ms
 3  2000:103::4 (2000:103::4)  0.013 ms  0.009 ms  0.009 ms

#!/bin/bash
#
#        ------------------------ N2
#         |                    |
#       ------              ------  N3  ----
#       | R1 |              | R2 |------|H2|
#       ------              ------      ----
#         |                    |
#        ------------------------ N1
#                  |
#                 ----
#                 |H1|
#                 ----
#
# N1: 2000:101::/64
# N2: 2000:102::/64
# N3: 2000:103::/64
#
# R1's host part of address: 1
# R2's host part of address: 2
# H1's host part of address: 3
# H2's host part of address: 4
#
# For example:
# the IPv6 address of R1's interface on N2 is 2000:102::1/64
#
# Nets are implemented by macvlan interfaces (bridge mode) over
# dummy interfaces.
#

# Create net namespaces
ip netns add host1
ip netns add host2
ip netns add rtr1
ip netns add rtr2

# Create nets
ip link add net1 type dummy; ip link set net1 up
ip link add net2 type dummy; ip link set net2 up
ip link add net3 type dummy; ip link set net3 up

# Add interfaces to net1, move them to their nemaspaces
ip link add link net1 dev host1net1 type macvlan mode bridge
ip link set host1net1 netns host1
ip link add link net1 dev rtr1net1 type macvlan mode bridge
ip link set rtr1net1 netns rtr1
ip link add link net1 dev rtr2net1 type macvlan mode bridge
ip link set rtr2net1 netns rtr2

# Add interfaces to net2, move them to their nemaspaces
ip link add link net2 dev rtr1net2 type macvlan mode bridge
ip link set rtr1net2 netns rtr1
ip link add link net2 dev rtr2net2 type macvlan mode bridge
ip link set rtr2net2 netns rtr2

# Add interfaces to net3, move them to their nemaspaces
ip link add link net3 dev rtr2net3 type macvlan mode bridge
ip link set rtr2net3 netns rtr2
ip link add link net3 dev host2net3 type macvlan mode bridge
ip link set host2net3 netns host2

# Configure interfaces and routes in host1
ip netns exec host1 ip link set lo up
ip netns exec host1 ip link set host1net1 up
ip netns exec host1 ip -6 addr add 2000:101::3/64 dev host1net1
ip netns exec host1 ip -6 route add default via 2000:101::1

# Configure interfaces and routes in rtr1
ip netns exec rtr1 ip link set lo up
ip netns exec rtr1 ip link set rtr1net1 up
ip netns exec rtr1 ip -6 addr add 2000:101::1/64 dev rtr1net1
ip netns exec rtr1 ip link set rtr1net2 up
ip netns exec rtr1 ip -6 addr add 2000:102::1/64 dev rtr1net2
ip netns exec rtr1 ip -6 route add default via 2000:102::2
ip netns exec rtr1 sysctl net.ipv6.conf.all.forwarding=1

# Configure interfaces and routes in rtr2
ip netns exec rtr2 ip link set lo up
ip netns exec rtr2 ip link set rtr2net1 up
ip netns exec rtr2 ip -6 addr add 2000:101::2/64 dev rtr2net1
ip netns exec rtr2 ip link set rtr2net2 up
ip netns exec rtr2 ip -6 addr add 2000:102::2/64 dev rtr2net2
ip netns exec rtr2 ip link set rtr2net3 up
ip netns exec rtr2 ip -6 addr add 2000:103::2/64 dev rtr2net3
ip netns exec rtr2 sysctl net.ipv6.conf.all.forwarding=1

# Configure interfaces and routes in host2
ip netns exec host2 ip link set lo up
ip netns exec host2 ip link set host2net3 up
ip netns exec host2 ip -6 addr add 2000:103::4/64 dev host2net3
ip netns exec host2 ip -6 route add default via 2000:103::2

# Ping host2 from host1
ip netns exec host1 ping6 -c5 2000:103::4

# Traceroute host2 from host1
ip netns exec host1 traceroute6 2000:103::4

# Delete nets
ip link del net3
ip link del net2
ip link del net1

# Delete namespaces
ip netns del rtr2
ip netns del rtr1
ip netns del host2
ip netns del host1

Signed-off-by: Francesco Ruggeri <fruggeri@arista.com>
Original-patch-by: Honggang Xu <hxu@arista.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotipc: improve message bundling algorithm
Tuong Lien [Fri, 1 Nov 2019 02:58:57 +0000 (09:58 +0700)]
tipc: improve message bundling algorithm

As mentioned in commit e95584a889e1 ("tipc: fix unlimited bundling of
small messages"), the current message bundling algorithm is inefficient
that can generate bundles of only one payload message, that causes
unnecessary overheads for both the sender and receiver.

This commit re-designs the 'tipc_msg_make_bundle()' function (now named
as 'tipc_msg_try_bundle()'), so that when a message comes at the first
place, we will just check & keep a reference to it if the message is
suitable for bundling. The message buffer will be put into the link
backlog queue and processed as normal. Later on, when another one comes
we will make a bundle with the first message if possible and so on...
This way, a bundle if really needed will always consist of at least two
payload messages. Otherwise, we let the first buffer go its way without
any need of bundling, so reduce the overheads to zero.

Moreover, since now we have both the messages in hand, we can even
optimize the 'tipc_msg_bundle()' function, make bundle of a very large
(size ~ MSS) and small messages which is not with the current algorithm
e.g. [1400-byte message] + [10-byte message] (MTU = 1500).

Acked-by: Ying Xue <ying.xue@windreiver.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: icmp: use input address in traceroute
Francesco Ruggeri [Fri, 1 Nov 2019 00:44:13 +0000 (17:44 -0700)]
net: icmp: use input address in traceroute

Even with icmp_errors_use_inbound_ifaddr set, traceroute returns the
primary address of the interface the packet was received on, even if
the path goes through a secondary address. In the example:

                    1.0.3.1/24
 ---- 1.0.1.3/24    1.0.1.1/24 ---- 1.0.2.1/24    1.0.2.4/24 ----
 |H1|--------------------------|R1|--------------------------|H2|
 ----            N1            ----            N2            ----

where 1.0.3.1/24 is R1's primary address on N1, traceroute from
H1 to H2 returns:

traceroute to 1.0.2.4 (1.0.2.4), 30 hops max, 60 byte packets
 1  1.0.3.1 (1.0.3.1)  0.018 ms  0.006 ms  0.006 ms
 2  1.0.2.4 (1.0.2.4)  0.021 ms  0.007 ms  0.007 ms

After applying this patch, it returns:

traceroute to 1.0.2.4 (1.0.2.4), 30 hops max, 60 byte packets
 1  1.0.1.1 (1.0.1.1)  0.033 ms  0.007 ms  0.006 ms
 2  1.0.2.4 (1.0.2.4)  0.011 ms  0.007 ms  0.007 ms

Original-patch-by: Bill Fenner <fenner@arista.com>
Signed-off-by: Francesco Ruggeri <fruggeri@arista.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'optimize-openvswitch-flow-looking-up'
David S. Miller [Mon, 4 Nov 2019 01:18:04 +0000 (17:18 -0800)]
Merge branch 'optimize-openvswitch-flow-looking-up'

Tonghao Zhang says:

====================
optimize openvswitch flow looking up

This series patch optimize openvswitch for performance or simplify
codes.

Patch 1, 2, 4: Port Pravin B Shelar patches to
linux upstream with little changes.

Patch 5, 6, 7: Optimize the flow looking up and
simplify the flow hash.

Patch 8, 9: are bugfix.

The performance test is on Intel Xeon E5-2630 v4.
The test topology is show as below:

+-----------------------------------+
|   +---------------------------+   |
|   | eth0   ovs-switch    eth1 |   | Host0
|   +---------------------------+   |
+-----------------------------------+
      ^                       |
      |                       |
      |                       |
      |                       |
      |                       v
+-----+----+             +----+-----+
| netperf  | Host1       | netserver| Host2
+----------+             +----------+

We use netperf send the 64B packets, and insert 255+ flow-mask:
$ ovs-dpctl add-flow ovs-switch "in_port(1),eth(dst=00:01:00:00:00:00/ff:ff:ff:ff:ff:01),eth_type(0x0800),ipv4(frag=no)" 2
...
$ ovs-dpctl add-flow ovs-switch "in_port(1),eth(dst=00:ff:00:00:00:00/ff:ff:ff:ff:ff:ff),eth_type(0x0800),ipv4(frag=no)" 2
$
$ netperf -t UDP_STREAM -H 2.2.2.200 -l 40 -- -m 18

* Without series patch, throughput 8.28Mbps
* With series patch, throughput 46.05Mbps

v6:
some coding style fixes

v5:
rewrite patch 8, release flow-mask when freeing flow

v4:
access ma->count with READ_ONCE/WRITE_ONCE API. More information,
see patch 5 comments.

v3:
update ma point when realloc mask_array in patch 5

v2:
simplify codes. e.g. use kfree_rcu instead of call_rcu
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: openvswitch: simplify the ovs_dp_cmd_new
Tonghao Zhang [Fri, 1 Nov 2019 14:23:54 +0000 (22:23 +0800)]
net: openvswitch: simplify the ovs_dp_cmd_new

use the specified functions to init resource.

Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Greg Rose <gvrose8192@gmail.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: openvswitch: don't unlock mutex when changing the user_features fails
Tonghao Zhang [Fri, 1 Nov 2019 14:23:53 +0000 (22:23 +0800)]
net: openvswitch: don't unlock mutex when changing the user_features fails

Unlocking of a not locked mutex is not allowed.
Other kernel thread may be in critical section while
we unlock it because of setting user_feature fail.

Fixes: 95a7233c4 ("net: openvswitch: Set OvS recirc_id from tc chain index")
Cc: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Greg Rose <gvrose8192@gmail.com>
Acked-by: William Tu <u9012063@gmail.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: openvswitch: fix possible memleak on destroy flow-table
Tonghao Zhang [Fri, 1 Nov 2019 14:23:52 +0000 (22:23 +0800)]
net: openvswitch: fix possible memleak on destroy flow-table

When we destroy the flow tables which may contain the flow_mask,
so release the flow mask struct.

Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Greg Rose <gvrose8192@gmail.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: openvswitch: add likely in flow_lookup
Tonghao Zhang [Fri, 1 Nov 2019 14:23:51 +0000 (22:23 +0800)]
net: openvswitch: add likely in flow_lookup

The most case *index < ma->max, and flow-mask is not NULL.
We add un/likely for performance.

Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Greg Rose <gvrose8192@gmail.com>
Acked-by: William Tu <u9012063@gmail.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: openvswitch: simplify the flow_hash
Tonghao Zhang [Fri, 1 Nov 2019 14:23:50 +0000 (22:23 +0800)]
net: openvswitch: simplify the flow_hash

Simplify the code and remove the unnecessary BUILD_BUG_ON.

Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Greg Rose <gvrose8192@gmail.com>
Acked-by: William Tu <u9012063@gmail.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: openvswitch: optimize flow-mask looking up
Tonghao Zhang [Fri, 1 Nov 2019 14:23:49 +0000 (22:23 +0800)]
net: openvswitch: optimize flow-mask looking up

The full looking up on flow table traverses all mask array.
If mask-array is too large, the number of invalid flow-mask
increase, performance will be drop.

One bad case, for example: M means flow-mask is valid and NULL
of flow-mask means deleted.

+-------------------------------------------+
| M | NULL | ...                  | NULL | M|
+-------------------------------------------+

In that case, without this patch, openvswitch will traverses all
mask array, because there will be one flow-mask in the tail. This
patch changes the way of flow-mask inserting and deleting, and the
mask array will be keep as below: there is not a NULL hole. In the
fast path, we can "break" "for" (not "continue") in flow_lookup
when we get a NULL flow-mask.

         "break"
            v
+-------------------------------------------+
| M | M |  NULL |...           | NULL | NULL|
+-------------------------------------------+

This patch don't optimize slow or control path, still using ma->max
to traverse. Slow path:
* tbl_mask_array_realloc
* ovs_flow_tbl_lookup_exact
* flow_mask_find

Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Greg Rose <gvrose8192@gmail.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: openvswitch: optimize flow mask cache hash collision
Tonghao Zhang [Fri, 1 Nov 2019 14:23:48 +0000 (22:23 +0800)]
net: openvswitch: optimize flow mask cache hash collision

Port the codes to linux upstream and with little changes.

Pravin B Shelar, says:
| In case hash collision on mask cache, OVS does extra flow
| lookup. Following patch avoid it.

Link: https://github.com/openvswitch/ovs/commit/0e6efbe2712da03522532dc5e84806a96f6a0dd1
Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Greg Rose <gvrose8192@gmail.com>
Signed-off-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: openvswitch: shrink the mask array if necessary
Tonghao Zhang [Fri, 1 Nov 2019 14:23:47 +0000 (22:23 +0800)]
net: openvswitch: shrink the mask array if necessary

When creating and inserting flow-mask, if there is no available
flow-mask, we realloc the mask array. When removing flow-mask,
if necessary, we shrink mask array.

Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Greg Rose <gvrose8192@gmail.com>
Acked-by: William Tu <u9012063@gmail.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: openvswitch: convert mask list in mask array
Tonghao Zhang [Fri, 1 Nov 2019 14:23:46 +0000 (22:23 +0800)]
net: openvswitch: convert mask list in mask array

Port the codes to linux upstream and with little changes.

Pravin B Shelar, says:
| mask caches index of mask in mask_list. On packet recv OVS
| need to traverse mask-list to get cached mask. Therefore array
| is better for retrieving cached mask. This also allows better
| cache replacement algorithm by directly checking mask's existence.

Link: https://github.com/openvswitch/ovs/commit/d49fc3ff53c65e4eca9cabd52ac63396746a7ef5
Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Greg Rose <gvrose8192@gmail.com>
Acked-by: William Tu <u9012063@gmail.com>
Signed-off-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: openvswitch: add flow-mask cache for performance
Tonghao Zhang [Fri, 1 Nov 2019 14:23:45 +0000 (22:23 +0800)]
net: openvswitch: add flow-mask cache for performance

The idea of this optimization comes from a patch which
is committed in 2014, openvswitch community. The author
is Pravin B Shelar. In order to get high performance, I
implement it again. Later patches will use it.

Pravin B Shelar, says:
| On every packet OVS needs to lookup flow-table with every
| mask until it finds a match. The packet flow-key is first
| masked with mask in the list and then the masked key is
| looked up in flow-table. Therefore number of masks can
| affect packet processing performance.

Link: https://github.com/openvswitch/ovs/commit/5604935e4e1cbc16611d2d97f50b717aa31e8ec5
Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Greg Rose <gvrose8192@gmail.com>
Acked-by: William Tu <u9012063@gmail.com>
Signed-off-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
David S. Miller [Sat, 2 Nov 2019 22:27:42 +0000 (15:27 -0700)]
Merge git://git./linux/kernel/git/bpf/bpf-next

Alexei Starovoitov says:

====================
pull-request: bpf-next 2019-11-02

The following pull-request contains BPF updates for your *net-next* tree.

We've added 30 non-merge commits during the last 7 day(s) which contain
a total of 41 files changed, 1864 insertions(+), 474 deletions(-).

The main changes are:

1) Fix long standing user vs kernel access issue by introducing
   bpf_probe_read_user() and bpf_probe_read_kernel() helpers, from Daniel.

2) Accelerated xskmap lookup, from Björn and Maciej.

3) Support for automatic map pinning in libbpf, from Toke.

4) Cleanup of BTF-enabled raw tracepoints, from Alexei.

5) Various fixes to libbpf and selftests.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
David S. Miller [Sat, 2 Nov 2019 20:12:51 +0000 (13:12 -0700)]
Merge git://git./linux/kernel/git/netdev/net

The only slightly tricky merge conflict was the netdevsim because the
mutex locking fix overlapped a lot of driver reload reorganization.

The rest were (relatively) trivial in nature.

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'bpf_probe_read_user'
Alexei Starovoitov [Sat, 2 Nov 2019 19:39:13 +0000 (12:39 -0700)]
Merge branch 'bpf_probe_read_user'

Daniel Borkmann says:

====================
This set adds probe_read_{user,kernel}(), probe_read_str_{user,kernel}()
helpers, fixes probe_write_user() helper and selftests. For details please
see individual patches.

Thanks!

v2 -> v3:
  - noticed two more things that are fixed in here:
   - bpf uapi helper description used 'int size' for *_str helpers, now u32
   - we need TASK_SIZE_MAX + guard page on x86-64 in patch 2 otherwise
     we'll trigger the 00c42373d397 warn as well, so full range covered now
v1 -> v2:
  - standardize unsafe_ptr terminology in uapi header comment (Andrii)
  - probe_read_{user,kernel}[_str] naming scheme (Andrii)
  - use global data in last test case, remove relaxed_maps (Andrii)
  - add strict non-pagefault kernel read funcs to avoid warning in
    kernel probe read helpers (Alexei)
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agobpf, testing: Add selftest to read/write sockaddr from user space
Daniel Borkmann [Fri, 1 Nov 2019 23:18:03 +0000 (00:18 +0100)]
bpf, testing: Add selftest to read/write sockaddr from user space

Tested on x86-64 and Ilya was also kind enough to give it a spin on
s390x, both passing with probe_user:OK there. The test is using the
newly added bpf_probe_read_user() to dump sockaddr from connect call
into .bss BPF map and overrides the user buffer via bpf_probe_write_user():

  # ./test_progs
  [...]
  #17 pkt_md_access:OK
  #18 probe_user:OK
  #19 prog_run_xattr:OK
  [...]

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/90f449d8af25354e05080e82fc6e2d3179da30ea.1572649915.git.daniel@iogearbox.net
5 years agobpf, testing: Convert prog tests to probe_read_{user, kernel}{, _str} helper
Daniel Borkmann [Fri, 1 Nov 2019 23:18:02 +0000 (00:18 +0100)]
bpf, testing: Convert prog tests to probe_read_{user, kernel}{, _str} helper

Use probe read *_{kernel,user}{,_str}() helpers instead of bpf_probe_read()
or bpf_probe_read_user_str() for program tests where appropriate.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/4a61d4b71ce3765587d8ef5cb93afa18515e5b3e.1572649915.git.daniel@iogearbox.net
5 years agobpf, samples: Use bpf_probe_read_user where appropriate
Daniel Borkmann [Fri, 1 Nov 2019 23:18:01 +0000 (00:18 +0100)]
bpf, samples: Use bpf_probe_read_user where appropriate

Use bpf_probe_read_user() helper instead of bpf_probe_read() for samples that
attach to kprobes probing on user addresses.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/5b0144b3f8e031ec5e2438bd7de8d7877e63bf2f.1572649915.git.daniel@iogearbox.net
5 years agobpf: Switch BPF probe insns to bpf_probe_read_kernel
Daniel Borkmann [Fri, 1 Nov 2019 23:18:00 +0000 (00:18 +0100)]
bpf: Switch BPF probe insns to bpf_probe_read_kernel

Commit 2a02759ef5f8 ("bpf: Add support for BTF pointers to interpreter")
explicitly states that the pointer to BTF object is a pointer to a kernel
object or NULL. Therefore we should also switch to using the strict kernel
probe helper which is restricted to kernel addresses only when architectures
have non-overlapping address spaces.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/d2b90827837685424a4b8008dfe0460558abfada.1572649915.git.daniel@iogearbox.net
5 years agobpf: Add probe_read_{user, kernel} and probe_read_{user, kernel}_str helpers
Daniel Borkmann [Fri, 1 Nov 2019 23:17:59 +0000 (00:17 +0100)]
bpf: Add probe_read_{user, kernel} and probe_read_{user, kernel}_str helpers

The current bpf_probe_read() and bpf_probe_read_str() helpers are broken
in that they assume they can be used for probing memory access for kernel
space addresses /as well as/ user space addresses.

However, plain use of probe_kernel_read() for both cases will attempt to
always access kernel space address space given access is performed under
KERNEL_DS and some archs in-fact have overlapping address spaces where a
kernel pointer and user pointer would have the /same/ address value and
therefore accessing application memory via bpf_probe_read{,_str}() would
read garbage values.

Lets fix BPF side by making use of recently added 3d7081822f7f ("uaccess:
Add non-pagefault user-space read functions"). Unfortunately, the only way
to fix this status quo is to add dedicated bpf_probe_read_{user,kernel}()
and bpf_probe_read_{user,kernel}_str() helpers. The bpf_probe_read{,_str}()
helpers are kept as-is to retain their current behavior.

The two *_user() variants attempt the access always under USER_DS set, the
two *_kernel() variants will -EFAULT when accessing user memory if the
underlying architecture has non-overlapping address ranges, also avoiding
throwing the kernel warning via 00c42373d397 ("x86-64: add warning for
non-canonical user access address dereferences").

Fixes: a5e8c07059d0 ("bpf: add bpf_probe_read_str helper")
Fixes: 2541517c32be ("tracing, perf: Implement BPF programs attached to kprobes")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/796ee46e948bc808d54891a1108435f8652c6ca4.1572649915.git.daniel@iogearbox.net
5 years agobpf: Make use of probe_user_write in probe write helper
Daniel Borkmann [Fri, 1 Nov 2019 23:17:58 +0000 (00:17 +0100)]
bpf: Make use of probe_user_write in probe write helper

Convert the bpf_probe_write_user() helper to probe_user_write() such that
writes are not attempted under KERNEL_DS anymore which is buggy as kernel
and user space pointers can have overlapping addresses. Also, given we have
the access_ok() check inside probe_user_write(), the helper doesn't need
to do it twice.

Fixes: 96ae52279594 ("bpf: Add bpf_probe_write_user BPF helper to be called in tracers")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/841c461781874c07a0ee404a454c3bc0459eed30.1572649915.git.daniel@iogearbox.net
5 years agouaccess: Add strict non-pagefault kernel-space read function
Daniel Borkmann [Fri, 1 Nov 2019 23:17:57 +0000 (00:17 +0100)]
uaccess: Add strict non-pagefault kernel-space read function

Add two new probe_kernel_read_strict() and strncpy_from_unsafe_strict()
helpers which by default alias to the __probe_kernel_read() and the
__strncpy_from_unsafe(), respectively, but can be overridden by archs
which have non-overlapping address ranges for kernel space and user
space in order to bail out with -EFAULT when attempting to probe user
memory including non-canonical user access addresses [0]:

  4-level page tables:
    user-space mem: 0x0000000000000000 - 0x00007fffffffffff
    non-canonical:  0x0000800000000000 - 0xffff7fffffffffff

  5-level page tables:
    user-space mem: 0x0000000000000000 - 0x00ffffffffffffff
    non-canonical:  0x0100000000000000 - 0xfeffffffffffffff

The idea is that these helpers are complementary to the probe_user_read()
and strncpy_from_unsafe_user() which probe user-only memory. Both added
helpers here do the same, but for kernel-only addresses.

Both set of helpers are going to be used for BPF tracing. They also
explicitly avoid throwing the splat for non-canonical user addresses from
00c42373d397 ("x86-64: add warning for non-canonical user access address
dereferences").

For compat, the current probe_kernel_read() and strncpy_from_unsafe() are
left as-is.

  [0] Documentation/x86/x86_64/mm.txt

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: x86@kernel.org
Link: https://lore.kernel.org/bpf/eefeefd769aa5a013531f491a71f0936779e916b.1572649915.git.daniel@iogearbox.net
5 years agouaccess: Add non-pagefault user-space write function
Daniel Borkmann [Fri, 1 Nov 2019 23:17:56 +0000 (00:17 +0100)]
uaccess: Add non-pagefault user-space write function

Commit 3d7081822f7f ("uaccess: Add non-pagefault user-space read functions")
missed to add probe write function, therefore factor out a probe_write_common()
helper with most logic of probe_kernel_write() except setting KERNEL_DS, and
add a new probe_user_write() helper so it can be used from BPF side.

Again, on some archs, the user address space and kernel address space can
co-exist and be overlapping, so in such case, setting KERNEL_DS would mean
that the given address is treated as being in kernel address space.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Link: https://lore.kernel.org/bpf/9df2542e68141bfa3addde631441ee45503856a8.1572649915.git.daniel@iogearbox.net
5 years agoMerge branch 'map-pinning'
Alexei Starovoitov [Sat, 2 Nov 2019 19:35:07 +0000 (12:35 -0700)]
Merge branch 'map-pinning'

Toke Høiland-Jørgensen says:

====================
This series adds support to libbpf for reading 'pinning' settings from BTF-based
map definitions. It introduces a new open option which can set the pinning path;
if no path is set, /sys/fs/bpf is used as the default. Callers can customise the
pinning between open and load by setting the pin path per map, and still get the
automatic reuse feature.

The semantics of the pinning is similar to the iproute2 "PIN_GLOBAL" setting,
and the eventual goal is to move the iproute2 implementation to be based on
libbpf and the functions introduced in this series.

Changelog:

v6:
  - Fix leak of struct bpf_object in selftest
  - Make struct bpf_map arg const in bpf_map__is_pinned() and bpf_map__get_pin_path()

v5:
  - Don't pin maps with pinning set, but with a value of LIBBPF_PIN_NONE
  - Add a few more selftests:
    - Should not pin map with pinning set, but value LIBBPF_PIN_NONE
    - Should fail to load a map with an invalid pinning value
    - Should fail to re-use maps with parameter mismatch
  - Alphabetise libbpf.map
  - Whitespace and typo fixes

v4:
  - Don't check key_type_id and value_type_id when checking for map reuse
    compatibility.
  - Move building of map->pin_path into init_user_btf_map()
  - Get rid of 'pinning' attribute in struct bpf_map
  - Make sure we also create parent directory on auto-pin (new patch 3).
  - Abort the selftest on error instead of attempting to continue.
  - Support unpinning all pinned maps with bpf_object__unpin_maps(obj, NULL)
  - Support pinning at map->pin_path with bpf_object__pin_maps(obj, NULL)
  - Make re-pinning a map at the same path a noop
  - Rename the open option to pin_root_path
  - Add a bunch more self-tests for pin_maps(NULL) and unpin_maps(NULL)
  - Fix a couple of smaller nits

v3:
  - Drop bpf_object__pin_maps_opts() and just use an open option to customise
    the pin path; also don't touch bpf_object__{un,}pin_maps()
  - Integrate pinning and reuse into bpf_object__create_maps() instead of having
    multiple loops though the map structure
  - Make errors in map reuse and pinning fatal to the load procedure
  - Add selftest to exercise pinning feature
  - Rebase series to latest bpf-next

v2:
  - Drop patch that adds mounting of bpffs
  - Only support a single value of the pinning attribute
  - Add patch to fixup error handling in reuse_fd()
  - Implement the full automatic pinning and map reuse logic on load
====================

Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoselftests: Add tests for automatic map pinning
Toke Høiland-Jørgensen [Sat, 2 Nov 2019 11:09:42 +0000 (12:09 +0100)]
selftests: Add tests for automatic map pinning

This adds a new BPF selftest to exercise the new automatic map pinning
code.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157269298209.394725.15420085139296213182.stgit@toke.dk
5 years agolibbpf: Add auto-pinning of maps when loading BPF objects
Toke Høiland-Jørgensen [Sat, 2 Nov 2019 11:09:41 +0000 (12:09 +0100)]
libbpf: Add auto-pinning of maps when loading BPF objects

This adds support to libbpf for setting map pinning information as part of
the BTF map declaration, to get automatic map pinning (and reuse) on load.
The pinning type currently only supports a single PIN_BY_NAME mode, where
each map will be pinned by its name in a path that can be overridden, but
defaults to /sys/fs/bpf.

Since auto-pinning only does something if any maps actually have a
'pinning' BTF attribute set, we default the new option to enabled, on the
assumption that seamless pinning is what most callers want.

When a map has a pin_path set at load time, libbpf will compare the map
pinned at that location (if any), and if the attributes match, will re-use
that map instead of creating a new one. If no existing map is found, the
newly created map will instead be pinned at the location.

Programs wanting to customise the pinning can override the pinning paths
using bpf_map__set_pin_path() before calling bpf_object__load() (including
setting it to NULL to disable pinning of a particular map).

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157269298092.394725.3966306029218559681.stgit@toke.dk
5 years agolibbpf: Move directory creation into _pin() functions
Toke Høiland-Jørgensen [Sat, 2 Nov 2019 11:09:39 +0000 (12:09 +0100)]
libbpf: Move directory creation into _pin() functions

The existing pin_*() functions all try to create the parent directory
before pinning. Move this check into the per-object _pin() functions
instead. This ensures consistent behaviour when auto-pinning is
added (which doesn't go through the top-level pin_maps() function), at the
cost of a few more calls to mkdir().

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157269297985.394725.5882630952992598610.stgit@toke.dk
5 years agolibbpf: Store map pin path and status in struct bpf_map
Toke Høiland-Jørgensen [Sat, 2 Nov 2019 11:09:38 +0000 (12:09 +0100)]
libbpf: Store map pin path and status in struct bpf_map

Support storing and setting a pin path in struct bpf_map, which can be used
for automatic pinning. Also store the pin status so we can avoid attempts
to re-pin a map that has already been pinned (or reused from a previous
pinning).

The behaviour of bpf_object__{un,}pin_maps() is changed so that if it is
called with a NULL path argument (which was previously illegal), it will
(un)pin only those maps that have a pin_path set.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157269297876.394725.14782206533681896279.stgit@toke.dk
5 years agolibbpf: Fix error handling in bpf_map__reuse_fd()
Toke Høiland-Jørgensen [Sat, 2 Nov 2019 11:09:37 +0000 (12:09 +0100)]
libbpf: Fix error handling in bpf_map__reuse_fd()

bpf_map__reuse_fd() was calling close() in the error path before returning
an error value based on errno. However, close can change errno, so that can
lead to potentially misleading error messages. Instead, explicitly store
errno in the err variable before each goto.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157269297769.394725.12634985106772698611.stgit@toke.dk
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Sat, 2 Nov 2019 00:48:11 +0000 (17:48 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Pull networking fixes from David Miller:

 1) Fix free/alloc races in batmanadv, from Sven Eckelmann.

 2) Several leaks and other fixes in kTLS support of mlx5 driver, from
    Tariq Toukan.

 3) BPF devmap_hash cost calculation can overflow on 32-bit, from Toke
    Høiland-Jørgensen.

 4) Add an r8152 device ID, from Kazutoshi Noguchi.

 5) Missing include in ipv6's addrconf.c, from Ben Dooks.

 6) Use siphash in flow dissector, from Eric Dumazet. Attackers can
    easily infer the 32-bit secret otherwise etc.

 7) Several netdevice nesting depth fixes from Taehee Yoo.

 8) Fix several KCSAN reported errors, from Eric Dumazet. For example,
    when doing lockless skb_queue_empty() checks, and accessing
    sk_napi_id/sk_incoming_cpu lockless as well.

 9) Fix jumbo packet handling in RXRPC, from David Howells.

10) Bump SOMAXCONN and tcp_max_syn_backlog values, from Eric Dumazet.

11) Fix DMA synchronization in gve driver, from Yangchun Fu.

12) Several bpf offload fixes, from Jakub Kicinski.

13) Fix sk_page_frag() recursion during memory reclaim, from Tejun Heo.

14) Fix ping latency during high traffic rates in hisilicon driver, from
    Jiangfent Xiao.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (146 commits)
  net: fix installing orphaned programs
  net: cls_bpf: fix NULL deref on offload filter removal
  selftests: bpf: Skip write only files in debugfs
  selftests: net: reuseport_dualstack: fix uninitalized parameter
  r8169: fix wrong PHY ID issue with RTL8168dp
  net: dsa: bcm_sf2: Fix IMP setup for port different than 8
  net: phylink: Fix phylink_dbg() macro
  gve: Fixes DMA synchronization.
  inet: stop leaking jiffies on the wire
  ixgbe: Remove duplicate clear_bit() call
  Documentation: networking: device drivers: Remove stray asterisks
  e1000: fix memory leaks
  i40e: Fix receive buffer starvation for AF_XDP
  igb: Fix constant media auto sense switching when no cable is connected
  net: ethernet: arc: add the missed clk_disable_unprepare
  igb: Enable media autosense for the i350.
  igb/igc: Don't warn on fatal read failures when the device is removed
  tcp: increase tcp_max_syn_backlog max value
  net: increase SOMAXCONN to 4096
  netdevsim: Fix use-after-free during device dismantle
  ...

5 years agoMerge tag 'nfs-for-5.4-3' of git://git.linux-nfs.org/projects/anna/linux-nfs
Linus Torvalds [Sat, 2 Nov 2019 00:37:44 +0000 (17:37 -0700)]
Merge tag 'nfs-for-5.4-3' of git://git.linux-nfs.org/projects/anna/linux-nfs

Pull NFS client bugfixes from Anna Schumaker:
 "This contains two delegation fixes (with the RCU lock leak fix marked
  for stable), and three patches to fix destroying the the sunrpc back
  channel.

  Stable bugfixes:

   - Fix an RCU lock leak in nfs4_refresh_delegation_stateid()

  Other fixes:

   - The TCP back channel mustn't disappear while requests are
     outstanding

   - The RDMA back channel mustn't disappear while requests are
     outstanding

   - Destroy the back channel when we destroy the host transport

   - Don't allow a cached open with a revoked delegation"

* tag 'nfs-for-5.4-3' of git://git.linux-nfs.org/projects/anna/linux-nfs:
  NFS: Fix an RCU lock leak in nfs4_refresh_delegation_stateid()
  NFSv4: Don't allow a cached open with a revoked delegation
  SUNRPC: Destroy the back channel when we destroy the host transport
  SUNRPC: The RDMA back channel mustn't disappear while requests are outstanding
  SUNRPC: The TCP back channel mustn't disappear while requests are outstanding

5 years agoMerge tag 'for-linus-20191101' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 2 Nov 2019 00:33:12 +0000 (17:33 -0700)]
Merge tag 'for-linus-20191101' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:

 - Two small nvme fixes, one is a fabrics connection fix, the other one
   a cleanup made possible by that fix (Anton, via Keith)

 - Fix requeue handling in umb ubd (Anton)

 - Fix spin_lock_irq() nesting in blk-iocost (Dan)

 - Three small io_uring fixes:
     - Install io_uring fd after done with ctx (me)
     - Clear ->result before every poll issue (me)
     - Fix leak of shadow request on error (Pavel)

* tag 'for-linus-20191101' of git://git.kernel.dk/linux-block:
  iocost: don't nest spin_lock_irq in ioc_weight_write()
  io_uring: ensure we clear io_kiocb->result before each issue
  um-ubd: Entrust re-queue to the upper layers
  nvme-multipath: remove unused groups_only mode in ana log
  nvme-multipath: fix possible io hang after ctrl reconnect
  io_uring: don't touch ctx in setup after ring fd install
  io_uring: Fix leaked shadow_req

5 years agoMerge tag 'riscv/for-v5.4-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv...
Linus Torvalds [Sat, 2 Nov 2019 00:20:53 +0000 (17:20 -0700)]
Merge tag 'riscv/for-v5.4-rc6' of git://git./linux/kernel/git/riscv/linux

Pull RISC-V fixes from Paul Walmsley:
 "One fix for PCIe users:

   - Fix legacy PCI I/O port access emulation

  One set of cleanups:

   - Resolve most of the warnings generated by sparse across arch/riscv.
     No functional changes

  And one MAINTAINERS update:

   - Update Palmer's E-mail address"

* tag 'riscv/for-v5.4-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux:
  MAINTAINERS: Change to my personal email address
  RISC-V: Add PCIe I/O BAR memory mapping
  riscv: for C functions called only from assembly, mark with __visible
  riscv: fp: add missing __user pointer annotations
  riscv: add missing header file includes
  riscv: mark some code and data as file-static
  riscv: init: merge split string literals in preprocessor directive
  riscv: add prototypes for assembly language functions from head.S

5 years agoMerge branch 'bpf-xskmap-perf-improvements'
Daniel Borkmann [Fri, 1 Nov 2019 23:38:49 +0000 (00:38 +0100)]
Merge branch 'bpf-xskmap-perf-improvements'

Björn Töpel says:

====================
This set consists of three patches from Maciej and myself which are
optimizing the XSKMAP lookups.  In the first patch, the sockets are
moved to be stored at the tail of the struct xsk_map. The second
patch, Maciej implements map_gen_lookup() for XSKMAP. The third patch,
introduced in this revision, moves various XSKMAP functions, to permit
the compiler to do more aggressive inlining.

Based on the XDP program from tools/lib/bpf/xsk.c where
bpf_map_lookup_elem() is explicitly called, this work yields a 5%
improvement for xdpsock's rxdrop scenario. The last patch yields 2%
improvement.

Jonathan's Acked-by: for patch 1 and 2 was carried on. Note that the
overflow checks are done in the bpf_map_area_alloc() and
bpf_map_charge_init() functions, which was fixed in commit
ff1c08e1f74b ("bpf: Change size to u64 for bpf_map_{area_alloc,
charge_init}()").

  [1] https://patchwork.ozlabs.org/patch/1186170/

v1->v2: * Change size/cost to size_t and use {struct, array}_size
          where appropriate. (Jakub)
v2->v3: * Proper commit message for patch 2.
v3->v4: * Change size_t to u64 to handle 32-bit overflows. (Jakub)
        * Introduced patch 3.
v4->v5: * Use BPF_SIZEOF size, instead of BPF_DW, for correct
          pointer-sized loads. (Daniel)
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoxsk: Restructure/inline XSKMAP lookup/redirect/flush
Björn Töpel [Fri, 1 Nov 2019 11:03:46 +0000 (12:03 +0100)]
xsk: Restructure/inline XSKMAP lookup/redirect/flush

In this commit the XSKMAP entry lookup function used by the XDP
redirect code is moved from the xskmap.c file to the xdp_sock.h
header, so the lookup can be inlined from, e.g., the
bpf_xdp_redirect_map() function.

Further the __xsk_map_redirect() and __xsk_map_flush() is moved to the
xsk.c, which lets the compiler inline the xsk_rcv() and xsk_flush()
functions.

Finally, all the XDP socket functions were moved from linux/bpf.h to
net/xdp_sock.h, where most of the XDP sockets functions are anyway.

This yields a ~2% performance boost for the xdpsock "rx_drop"
scenario.

Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20191101110346.15004-4-bjorn.topel@gmail.com
5 years agobpf: Implement map_gen_lookup() callback for XSKMAP
Maciej Fijalkowski [Fri, 1 Nov 2019 11:03:45 +0000 (12:03 +0100)]
bpf: Implement map_gen_lookup() callback for XSKMAP

Inline the xsk_map_lookup_elem() via implementing the map_gen_lookup()
callback. This results in emitting the bpf instructions in place of
bpf_map_lookup_elem() helper call and better performance of bpf
programs.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Link: https://lore.kernel.org/bpf/20191101110346.15004-3-bjorn.topel@gmail.com
5 years agoxsk: Store struct xdp_sock as a flexible array member of the XSKMAP
Björn Töpel [Fri, 1 Nov 2019 11:03:44 +0000 (12:03 +0100)]
xsk: Store struct xdp_sock as a flexible array member of the XSKMAP

Prior this commit, the array storing XDP socket instances were stored
in a separate allocated array of the XSKMAP. Now, we store the sockets
as a flexible array member in a similar fashion as the arraymap. Doing
so, we do less pointer chasing in the lookup.

Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Link: https://lore.kernel.org/bpf/20191101110346.15004-2-bjorn.topel@gmail.com
5 years agoMerge branch 'parisc-5.4-3' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Fri, 1 Nov 2019 22:16:25 +0000 (15:16 -0700)]
Merge branch 'parisc-5.4-3' of git://git./linux/kernel/git/deller/parisc-linux

Pull parisc fix from Helge Deller:
 "Fix a parisc kernel crash with ftrace functions when compiled without
  frame pointers"

* 'parisc-5.4-3' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: fix frame pointer in ftrace_regs_caller()

5 years agoMerge branch 'fix-BPF-offload-related-bugs'
David S. Miller [Fri, 1 Nov 2019 22:16:01 +0000 (15:16 -0700)]
Merge branch 'fix-BPF-offload-related-bugs'

Jakub Kicinski says:

====================
fix BPF offload related bugs

test_offload.py catches some recently added bugs.

First of a bug in test_offload.py itself after recent changes
to netdevsim is fixed.

Second patch fixes a bug in cls_bpf, and last one addresses
a problem with the recently added XDP installation optimization.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: fix installing orphaned programs
Jakub Kicinski [Fri, 1 Nov 2019 03:07:00 +0000 (20:07 -0700)]
net: fix installing orphaned programs

When netdevice with offloaded BPF programs is destroyed
the programs are orphaned and removed from the program
IDA - their IDs get released (the programs may remain
accessible via existing open file descriptors and pinned
files). After IDs are released they are set to 0.

This confuses dev_change_xdp_fd() because it compares
the __dev_xdp_query() result where 0 means no program
with prog->aux->id where 0 means orphaned.

dev_change_xdp_fd() would have incorrectly returned success
even though it had not installed the program.

Since drivers already catch this case via bpf_offload_dev_match()
let them handle this case. The error message drivers produce in
this case ("program loaded for a different device") is in fact
correct as the orphaned program must had to be loaded for a
different device.

Fixes: c14a9f633d9e ("net: Don't call XDP_SETUP_PROG when nothing is changed")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: cls_bpf: fix NULL deref on offload filter removal
Jakub Kicinski [Fri, 1 Nov 2019 03:06:59 +0000 (20:06 -0700)]
net: cls_bpf: fix NULL deref on offload filter removal

Commit 401192113730 ("net: sched: refactor block offloads counter
usage") missed the fact that either new prog or old prog may be
NULL.

Fixes: 401192113730 ("net: sched: refactor block offloads counter usage")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>