Tariq Toukan says:

====================
mlx4 XDP TX refactor

This patchset refactors the XDP forwarding case, so that
its dedicated transmit queues are managed in a complete
separation from the other regular ones.

It also adds ethtool counters for XDP cases.

Series generated against net-next commit:
22ca904a genetlink: fix error return code in genl_register_family()

Thanks,
Tariq.

v3:
* Exposed per ring counters.

v2:
* Added ethtool counters.
* Rebased, now patch 2 reverts Brenden's fix, as the bug no longer exists:
  958b3d39 ("net/mlx4_en: fixup xdp tx irq to match rx")
* Updated commit message of patch 2.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

XDP statistics are reported in ethtool, in total and per ring,
as follows:
- xdp_drop: the number of packets dropped by xdp.
- xdp_tx: the number of packets forwarded by xdp.
- xdp_tx_full: the number of times an xdp forward failed
	due to a full tx xdp ring.

In addition, all packets that are dropped/forwarded by XDP
are no longer accounted in rx_packets/rx_bytes of the ring,
so that they count traffic that is passed to the stack.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Separately manage the two types of TX rings: regular ones, and XDP.
Upon an XDP set, do not borrow regular TX rings and convert them
into XDP ones, but allocate new ones, unless we hit the max number
of rings.
Which means that in systems with smaller #cores we will not consume
the current TX rings for XDP, while we are still in the num TX limit.

XDP TX rings counters are not shown in ethtool statistics.
Instead, XDP counters will be added to the respective RX rings
in a downstream patch.

This has no performance implications.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Support XDP CQ type, and refactor the CQ type enum.
Rename the is_tx field to match the change.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

After adding sctp gso, sctp_packet_transmit is a quite big function now.

This patch is to extract the codes for packing packet to sctp_packet_pack
from sctp_packet_transmit, and add some comments, simplify the err path by
freeing auth chunk when freeing packet chunk_list in out path and freeing
head skb early if it fails to pack packet.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Roi Dayan says:

====================
misc TC/flower changes

This series includes two small changes to the TC flower classifier.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Move common code from fl_delete and fl_detroy to __fl_delete.
Signed-off-by: Roi Dayan <roid@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcf_unbind was called in fl_delete but was missing in fl_destroy when
force deleting flows.

Fixes: 77b9900e ('tc: introduce Flower classifier')
Signed-off-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Pablo Neira Ayuso says:

====================
Netfilter updates for net-next

The following patchset contains Netfilter updates for your net-next
tree. This includes better integration with the routing subsystem for
nf_tables, explicit notrack support and smaller updates. More
specifically, they are:

1) Add fib lookup expression for nf_tables, from Florian Westphal. This
   new expression provides a native replacement for iptables addrtype
   and rp_filter matches. This is more flexible though, since we can
   populate the kernel flowi representation to inquire fib to
   accomodate new usecases, such as RTBH through skb mark.

2) Introduce rt expression for nf_tables, from Anders K. Pedersen. This
   new expression allow you to access skbuff route metadata, more
   specifically nexthop and classid fields.

3) Add notrack support for nf_tables, to skip conntracking, requested by
   many users already.

4) Add boilerplate code to allow to use nf_log infrastructure from
   nf_tables ingress.

5) Allow to mangle pkttype from nf_tables prerouting chain, to emulate
   the xtables cluster match, from Liping Zhang.

6) Move socket lookup code into generic nf_socket_* infrastructure so
   we can provide a native replacement for the xtables socket match.

7) Make sure nfnetlink_queue data that is updated on every packets is
   placed in a different cache from read-only data, from Florian Westphal.

8) Handle NF_STOLEN from nf_tables core, also from Florian Westphal.

9) Start round robin number generation in nft_numgen from zero,
   instead of n-1, for consistency with xtables statistics match,
   patch from Liping Zhang.

10) Set GFP_NOWARN flag in skbuff netlink allocations in nfnetlink_log,
    given we retry with a smaller allocation on failure, from Calvin Owens.

11) Cleanup xt_multiport to use switch(), from Gao feng.

12) Remove superfluous check in nft_immediate and nft_cmp, from
    Liping Zhang.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

As the comment indicates, the data at the end of nfqnl_instance struct is
written on every queue/dequeue, so it should reside in its own cacheline.

Before this change, 'lock' was in first cacheline so we dirtied both.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

After call nft_data_init, size is already validated and desc.len will
not exceed the sizeof(struct nft_data), i.e. 16 bytes. So it will never
exceed U8_MAX.

Furthermore, in nft_immediate_init, we forget to call nft_data_uninit
when desc.len exceeds U8_MAX, although this will not happen, but it's
a logical mistake.

Now remove these redundant validation introduced by commit 36b701fa
("netfilter: nf_tables: validate maximum value of u32 netlink attributes")
Signed-off-by: Liping Zhang <zlpnobody@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Introduces an nftables rt expression for routing related data with support
for nexthop (i.e. the directly connected IP address that an outgoing packet
is sent to), which can be used either for matching or accounting, eg.

 # nft add rule filter postrouting \
	ip daddr 192.168.1.0/24 rt nexthop != 192.168.0.1 drop

This will drop any traffic to 192.168.1.0/24 that is not routed via
192.168.0.1.

 # nft add rule filter postrouting \
	flow table acct { rt nexthop timeout 600s counter }
 # nft add rule ip6 filter postrouting \
	flow table acct { rt nexthop timeout 600s counter }

These rules count outgoing traffic per nexthop. Note that the timeout
releases an entry if no traffic is seen for this nexthop within 10 minutes.

 # nft add rule inet filter postrouting \
	ether type ip \
	flow table acct { rt nexthop timeout 600s counter }
 # nft add rule inet filter postrouting \
	ether type ip6 \
	flow table acct { rt nexthop timeout 600s counter }

Same as above, but via the inet family, where the ether type must be
specified explicitly.

"rt classid" is also implemented identical to "meta rtclassid", since it
is more logical to have this match in the routing expression going forward.
Signed-off-by: Anders K. Pedersen <akp@cohaesio.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

We need this split to reuse existing codebase for the upcoming nf_tables
socket expression.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Move layer 2 packet logging into nf_log_l2packet() that resides in
nf_log_common.c, so this can be shared by both bridge and netdev
families.

This patch adds the boiler plate code to register the netdev logging
family.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Add FIB expression, supported for ipv4, ipv6 and inet family (the latter
just dispatches to ipv4 or ipv6 one based on nfproto).

Currently supports fetching output interface index/name and the
rtm_type associated with an address.

This can be used for adding path filtering. rtm_type is useful
to e.g. enforce a strong-end host model where packets
are only accepted if daddr is configured on the interface the
packet arrived on.

The fib expression is a native nftables alternative to the
xtables addrtype and rp_filter matches.

FIB result order for oif/oifname retrieval is as follows:
 - if packet is local (skb has rtable, RTF_LOCAL set, this
   will also catch looped-back multicast packets), set oif to
   the loopback interface.
 - if fib lookup returns an error, or result points to local,
   store zero result.  This means '--local' option of -m rpfilter
   is not supported. It is possible to use 'fib type local' or add
   explicit saddr/daddr matching rules to create exceptions if this
   is really needed.
 - store result in the destination register.
   In case of multiple routes, search set for desired oif in case
   strict matching is requested.

ipv4 and ipv6 behave fib expressions are supposed to behave the same.

[ I have collapsed Arnd Bergmann's ("netfilter: nf_tables: fib warnings")

	http://patchwork.ozlabs.org/patch/688615/

  to address fallout from this patch after rebasing nf-next, that was
  posted to address compilation warnings. --pablo ]
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

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

Also fix the return value check of idr_alloc() since idr_alloc return
negative errors on failure, not zero.

Fixes: 2ae0f17d ("genetlink: use idr to track families")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Enable support for IPv4 multicast:
- similar to unicast the flow struct is updated to L3 master device
  if relevant prior to calling fib_rules_lookup. The table id is saved
  to the lookup arg so the rule action for ipmr can return the table
  associated with the device.

- ip_mr_forward needs to check for master device mismatch as well
  since the skb->dev is set to it

- allow multicast address on VRF device for Rx by checking for the
  daddr in the VRF device as well as the original ingress device

- on Tx need to drop to __mkroute_output when FIB lookup fails for
  multicast destination address.

- if CONFIG_IP_MROUTE_MULTIPLE_TABLES is enabled VRF driver creates
  IPMR FIB rules on first device create similar to FIB rules. In
  addition the VRF driver does not divert IPv4 multicast packets:
  it breaks on Tx since the fib lookup fails on the mcast address.

With this patch, ipmr forwarding and local rx/tx work.
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Parthasarathy Bhuvaragan says:

====================
tipc: socket layer improvements

The following issues with the current socket layer hinders socket diagnostics
implementation, which led to this patch series.

1. tipc socket state is derived from multiple variables like
   sock->state, tsk->probing_state and tsk->connected. This style forces
   us to export multiple attributes to the user space, which has to be
   backward compatible.

2. Abuse of sock->state cannot be exported to user-space without
   requiring tipc specific hacks in the user-space.
   - For connection less (CL) sockets sock->state is overloaded to
     tipc state SS_READY.
   - For connection oriented (CO) listening socket sock->state is
     overloaded to tipc state SS_LISTEN.

This series is split into four:
1. Bug fixes in patch #1,2,3.
2. Minor cleanups in patch#4-5.
3. Express all tipc states using a single variable in patch#6-8.
4. Migrate the new tipc states to sk->sk_state in patch#9-16.

The figures below represents the FSM after this series:

Stream Server Listening Socket:
+-----------+       +-------------+
| TIPC_OPEN |------>| TIPC_LISTEN |
+-----------+       +-------------+

Stream Server Data Socket:
+-----------+       +------------------+
| TIPC_OPEN |------>| TIPC_ESTABLISHED |
+-----------+       +------------------+
                          ^   |
                          |   |
                          |   v
                    +--------------------+
                    | TIPC_DISCONNECTING |
                    +--------------------+

Stream Socket Client:
+-----------+       +-----------------+
| TIPC_OPEN |------>| TIPC_CONNECTING |------+
+-----------+       +-----------------+      |
                            |                |
                            |                |
                            v                |
                    +------------------+     |
                    | TIPC_ESTABLISHED |     |
                    +------------------+     |
                          ^   |              |
                          |   |              |
                          |   v              |
                    +--------------------+   |
                    | TIPC_DISCONNECTING |<--+
                    +--------------------+

NOTE:
This is just a base refractoring required for socket diagnostics.
TIPC socket diagnostics support will be introduced in a later series.

v2: - remove extra cast and parenthesis as suggested by David S. Miller in #4.
    - map new tipc state values to tcp states to address Eric Dumazet's concern,
      thus allow the usage of generic sk_* helpers. This is done in patch#10-15.
    - remove TIPC_PROBING state and replace it with probe_unacked flag in #11.
    - replace the TIPC_CLOSING state in v1 with sk_shutdown flag in #14.
    - introduce __tipc_shutdown() to avoid code duplication in #14.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

In this commit, we replace references to sock->state SS_CONNECTE
with sk_state TIPC_ESTABLISHED.

Finally, the sock->state is no longer explicitly used by tipc.
The FSM below is for various types of connection oriented sockets.

Stream Server Listening Socket:
+-----------+       +-------------+
| TIPC_OPEN |------>| TIPC_LISTEN |
+-----------+       +-------------+

Stream Server Data Socket:
+-----------+       +------------------+
| TIPC_OPEN |------>| TIPC_ESTABLISHED |
+-----------+       +------------------+
                          ^   |
                          |   |
                          |   v
                    +--------------------+
                    | TIPC_DISCONNECTING |
                    +--------------------+

Stream Socket Client:
+-----------+       +-----------------+
| TIPC_OPEN |------>| TIPC_CONNECTING |------+
+-----------+       +-----------------+      |
                            |                |
                            |                |
                            v                |
                    +------------------+     |
                    | TIPC_ESTABLISHED |     |
                    +------------------+     |
                          ^   |              |
                          |   |              |
                          |   v              |
                    +--------------------+   |
                    | TIPC_DISCONNECTING |<--+
                    +--------------------+
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In this commit, we create a new tipc socket state TIPC_CONNECTING
by primarily replacing the SS_CONNECTING with TIPC_CONNECTING.

There is no functional change in this commit.
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In this commit, we replace the references to SS_DISCONNECTING with
the combination of sk_state TIPC_DISCONNECTING and flags set in
sk_shutdown.
We introduce a new function _tipc_shutdown(), which provides
the common code required by tipc_release() and tipc_shutdown().
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In this commit, we create a new tipc socket state TIPC_DISCONNECTING in
sk_state. TIPC_DISCONNECTING is replacing the socket connection status
update using SS_DISCONNECTING.
TIPC_DISCONNECTING is set for connection oriented sockets at:
- tipc_shutdown()
- connection probe timeout
- when we receive an error message on the connection.

There is no functional change in this commit.
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In this commit, we create a new tipc socket state TIPC_OPEN in
sk_state. We primarily replace the SS_UNCONNECTED sock->state with
TIPC_OPEN.
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Until now, tipc maintains probing state for connected sockets in
tsk->probing_state variable.

In this commit, we express this information as socket states and
this remove the variable. We set probe_unacked flag when a probe
is sent out and reset it if we receive a reply. Instead of the
probing state TIPC_CONN_OK, we create a new state TIPC_ESTABLISHED.

There is no functional change in this commit.
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Until now, tipc maintains the socket state in sock->state variable.
This is used to maintain generic socket states, but in tipc
we overload it and save tipc socket states like TIPC_LISTEN.
Other protocols like TCP, UDP store protocol specific states
in sk->sk_state instead.

In this commit, we :
- declare a new tipc state TIPC_LISTEN, that replaces SS_LISTEN
- Create a new function tipc_set_state(), to update sk->sk_state.
- TIPC_LISTEN state is maintained in sk->sk_state.
- replace references to SS_LISTEN with TIPC_LISTEN.

There is no functional change in this commit.
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Until now, tipc socket state SS_READY declares that the socket is a
connectionless socket.

In this commit, we remove the state SS_READY and replace it with a
condition which returns true for datagram / connectionless sockets.
Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Until now, probing_intv is a variable in struct tipc_sock but is
always set to a constant CONN_PROBING_INTERVAL. The socket
connection is probed based on this value.

In this commit, we remove this variable and setup the socket
timer based on the constant CONN_PROBING_INTERVAL.

There is no functional change in this commit.
Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Until now, we determine if a socket is connected or not based on
tsk->connected, which is set once when the probing state is set
to TIPC_CONN_OK. It is unset when the sock->state is updated from
SS_CONNECTED to any other state.

In this commit, we remove connected variable from tipc_sock and
derive socket connection status from the following condition:
sock->state == SS_CONNECTED => tsk->connected

There is no functional change in this commit.
Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Until now, for connectionless sockets the peer information during
connect is stored in tsk->peer and a connection state is set in
tsk->connected. This is redundant.

In this commit, for connectionless sockets we update:
- __tipc_sendmsg(), when the destination is NULL the peer existence
  is determined by tsk->peer.family, instead of tsk->connected.
- tipc_connect(), remove set/unset of tsk->connected.
Hence tsk->connected is no longer used for connectionless sockets.

There is no functional change in this commit.
Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Until now, the peer information for connect is stored in tsk->remote
but the rest of code uses the name peer for peer/remote.

In this commit, we rename tsk->remote to tsk->peer to align with
naming convention followed in the rest of the code.

There is no functional change in this commit.
Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In this commit, we rename handle to bytes_read indicating the
purpose of the member.
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Until now, tipc_accept() calls sk_alloc() with kern=1. This is
incorrect as the data socket's owner is the user application.
Thus for these accepted data sockets the network namespace
refcount is skipped.

In this commit, we fix this by setting kern=0.
Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Until now, in filter_connect() when we terminate a connection due to
an error message from peer, we set the socket state to DISCONNECTING.

The socket is notified about this broken connection using EPIPE when
a user tries to send a message. However if a socket was waiting on a
poll() while the connection is being terminated, we fail to wakeup
that socket.

In this commit, we wakeup sleeping sockets at connection termination.
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Until now, in stream/mcast send() we pass the message to the link
layer even when the link is congested and add the socket to the
link's wakeup queue. This is unnecessary for non-blocking sockets.
If a socket is set to non-blocking and sends multicast with zero
back off time while receiving EAGAIN, we exhaust the memory.

In this commit, we return immediately at stream/mcast send() for
non-blocking sockets.
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Jakub Kicinski says:

====================
nfp: cleanups and RX path rewrite

This series lays groundwork for upcoming XDP support by updating
the RX path not to pre-allocate sk_buffs.  I start with few
cleanups, removal of NFP3200-related code being the most significant.
Patch 7 moves to alloc_frag() and build_skb() APIs.  Again, a number
of small cleanups follow.  The set ends with adding support for
different number of RX and TX rings.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

We used to always allocate the same number of TX and RX rings
so the support for having r_vectors without one of the rings
was dropped.  That makes us, however, unnecessarily limited
to 8 TX rings (8 is the Linux RSS default) most of the time.
Also we are about to add channel count configuration via
ethtool, so bring that support back.  TX rings can now default
to num_online_cpus() and RX rings to netif_get_num_default_rss_queues().
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

num_irqs is not used anywhere, replace it with max_r_vecs which holds
number of allocated RX/TX vectors and is going to be useful soon.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

nfp_net_irqs_wanted() doesn't really encapsulate much logic,
remove it and inline the calculations.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Use unsigned int consistently for vector/ring counts.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

We are currently using define for max TX rings to allocate IRQ
vectors.  It's OK since the max number of rings for TX and RX
are currently the same, but lets make the code nicer by taking
max of the two.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>