soreuseport: fast reuseport TCP socket selection

This change extends the fast SO_REUSEPORT socket lookup implemented
for UDP to TCP.  Listener sockets with SO_REUSEPORT and the same
receive address are additionally added to an array for faster
random access.  This means that only a single socket from the group
must be found in the listener list before any socket in the group can
be used to receive a packet.  Previously, every socket in the group
needed to be considered before handing off the incoming packet.

This feature also exposes the ability to use a BPF program when
selecting a socket from a reuseport group.

Signed-off-by: Craig Gallek <kraig@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

soreuseport: Prep for fast reuseport TCP socket selection

Both of the lines in this patch probably should have been included
in the initial implementation of this code for generic socket
support, but weren't technically necessary since only UDP sockets
were supported.

First, the sk_reuseport_cb points to a structure which assumes
each socket in the group has this pointer assigned at the same
time it's added to the array in the structure.  The sk_clone_lock
function breaks this assumption.  Since a child socket shouldn't
implicitly be in a reuseport group, the simple fix is to clear
the field in the clone.

Second, the SO_ATTACH_REUSEPORT_xBPF socket options require that
SO_REUSEPORT also be set first.  For UDP sockets, this is easily
enforced at bind-time since that process both puts the socket in
the appropriate receive hlist and updates the reuseport structures.
Since these operations can happen at two different times for TCP
sockets (bind and listen) it must be explicitly checked to enforce
the use of SO_REUSEPORT with SO_ATTACH_REUSEPORT_xBPF in the
setsockopt call.

Signed-off-by: Craig Gallek <kraig@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

inet: refactor inet[6]_lookup functions to take skb

This is a preliminary step to allow fast socket lookup of SO_REUSEPORT
groups.  Doing so with a BPF filter will require access to the
skb in question.  This change plumbs the skb (and offset to payload
data) through the call stack to the listening socket lookup
implementations where it will be used in a following patch.

Signed-off-by: Craig Gallek <kraig@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: __tcp_hdrlen() helper

tcp_hdrlen is wasteful if you already have a pointer to struct tcphdr.
This splits the size calculation into a helper function that can be
used if a struct tcphdr is already available.

Signed-off-by: Craig Gallek <kraig@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

inet: create IPv6-equivalent inet_hash function

In order to support fast lookups for TCP sockets with SO_REUSEPORT,
the function that adds sockets to the listening hash set needs
to be able to check receive address equality.  Since this equality
check is different for IPv4 and IPv6, we will need two different
socket hashing functions.

This patch adds inet6_hash identical to the existing inet_hash function
and updates the appropriate references.  A following patch will
differentiate the two by passing different comparison functions to
__inet_hash.

Additionally, in order to use the IPv6 address equality function from
inet6_hashtables (which is compiled as a built-in object when IPv6 is
enabled) it also needs to be in a built-in object file as well.  This
moves ipv6_rcv_saddr_equal into inet_hashtables to accomplish this.

Signed-off-by: Craig Gallek <kraig@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sock: struct proto hash function may error

In order to support fast reuseport lookups in TCP, the hash function
defined in struct proto must be capable of returning an error code.
This patch changes the function signature of all related hash functions
to return an integer and handles or propagates this return value at
all call sites.

Signed-off-by: Craig Gallek <kraig@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'batman-adv-for-davem' of git://git.open-mesh.org/linux-merge

Antonio Quartulli says:

====================
Here you have a batch of patches by Sven Eckelmann that
drops our private reference counting implementation and
substitutes it with the kref objects/functions.

Then you have a patch, by Simon Wunderlich, that
makes the broadcast protection window code more generic so
that it can be re-used in the future by other components
with different requirements.

Lastly, Sven is also introducing two lockdep asserts in
functions operating on our TVLV container list, to make
sure that the proper lock is always acquired by the users.
====================

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

Merge branch 'be2net-next'

Ajit Khaparde says:

====================
be2net Patch series

Please consider applying these two patches to net-next

  Patch-1: Request RSS capability of Rx interface depending on number of
    Rx rings
  Patch-2: Interpret and log new data that's added to the port
    misconfigure async event
====================

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

be2net: Interpret and log new data that's added to the port misconfigure async event

>From FW version 11.0. onwards, the PORT_MISCONFIG event generated by the FW
will carry more information about the event in the "data_word1"
and "data_word2" fields. This patch adds support in the driver to parse the
new information and log it accordingly. This patch also changes some of the
messages that are being logged currently.

Signed-off-by: Suresh Reddy <suresh.reddy@broadcom.com>
Signed-off-by: Venkat Duvvuru <venkatkumar.duvvuru@broadcom.com>
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: Request RSS capability of Rx interface depending on number of Rx rings

Currently we request RSS capability even if a single Rx ring is created.
As a result in few cases we unnecessarily consume an RSS capable interface
which is a limited resource in the chip.
This patch enables RSS on an interface only if more than one Rx ring
is created.

Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

batman-adv: Convert batadv_tt_common_entry to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_orig_node to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_orig_node_vlan to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_hard_iface to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_neigh_node to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_orig_ifinfo to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_neigh_ifinfo to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_tt_orig_list_entry to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_tvlv_handler to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_tvlv_container to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_dat_entry to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_nc_path to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_nc_node to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_bla_claim to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_bla_backbone_gw to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_softif_vlan to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_gw_node to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Convert batadv_hardif_neigh_node to kref

batman-adv uses a self-written reference implementation which is just based
on atomic_t. This is less obvious when reading the code than kref and
therefore increases the change that the reference counting will be missed.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Add lockdep assert for container_list_lock

The batadv_tvlv_container* functions state in their kernel-doc that they
require tvlv.container_list_lock. Add an assert to automatically detect
when this might have been ignored by the caller.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: add seqno maximum age and protection start flag parameters

To allow future use of the window protected function with different
maximum sequence numbers, add a parameter to set this value which
was previously hardcoded. Another parameter added for future use is a
flag to return whether the protection window has started.

While at it, also fix the kerneldoc.

Signed-off-by: Simon Wunderlich <simon@open-mesh.com>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

batman-adv: Drop reference to netdevice on last reference

The references to the network device should be dropped inside the release
function for batadv_hard_iface similar to what is done with the batman-adv
internal datastructures.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
Signed-off-by: Antonio Quartulli <a@unstable.cc>

sxgbe: remove unused code

Remove the unused code of sxgbe_xpcs.

Reported-by: Julia Lawall <julia.lawall@lip6.fr>
Suggested-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Jean Sacren <sakiwit@gmail.com>
Cc: Byungho An <bh74.an@samsung.com>
Cc: Girish K S <ks.giri@samsung.com>
Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1601191918470.2531@hadrien
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'renesas-bit-twiddling'

Sergei Shtylyov says:

====================
Factor out register bit twiddling in the Renesas Ethernet drivers

   Here's a set of 2 patches against DaveM's 'net-next.git' repo. We factor out
the often repeated pattern of reading a register, AND'ing and/or OR'ing some
bits, and then writing the value back.

[1/2] ravb: factor out register bit twiddling code
[2/2] sh_eth: factor out register bit twiddling code
====================

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

sh_eth: factor out register bit twiddling code

The  driver has often repeated pattern of reading a register,  AND'ing and/or
OR'ing some bits  and writing  the  value back. Factor the pattern out into
sh_eth_modify() -- this saves  84 bytes of code with ARM gcc 4.7.3.

While at it, update Cogent Embedded's copyright.

Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ravb: factor out register bit twiddling code

The  driver has often repeated pattern of reading a register,  AND'ing and/or
OR'ing some bits  and writing  the  value back. Factor the pattern out into
ravb_modify() -- this saves 260 bytes of code with ARM gcc 4.7.3.

While at it, update Cogent Embedded's copyrights.

Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'tpacket-gso-csum-offload'

Willem de Bruijn says:

====================
packet: tpacket gso and csum offload

Extend PACKET_VNET_HDR socket option support to packet sockets with
memory mapped rings.

Patches 2 and 4 add support to tpacket_rcv and tpacket_snd.

Patch 1 prepares for this by moving the relevant virtio_net_hdr
logic out of packet_snd and packet_rcv into helper functions.

GSO transmission requires all headers in the skb linear section.
Patch 3 moves parsing of tx_ring slot headers before skb allocation
to enable allocation with sufficient linear size.

Changes
  v1->v2:
    - fix bounds checks:
      - subtract sizeof(vnet_hdr) before comparing tp_len to size_max
      - compare tp_len to size_max also with GSO, just do not truncate to MTU
====================

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

packet: tpacket_snd gso and checksum offload

Support socket option PACKET_VNET_HDR together with PACKET_TX_RING.

When enabled, a struct virtio_net_hdr is expected to precede the data
in the ring. The vnet option must be set before the ring is created.

The implementation reuses the existing skb_copy_bits code that is used
when dev->hard_header_len is non-zero. Move this ll_header check to
before the skb alloc and combine it with a test for vnet_hdr->hdr_len.
Allocate and copy the max of the two.

Verified with test program at
github.com/wdebruij/kerneltools/blob/master/tests/psock_txring_vnet.c

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

packet: parse tpacket header before skb alloc

GSO packet headers must be stored in the linear skb segment.
Move tpacket header parsing before sock_alloc_send_skb. The GSO
follow-on patch will later increase the skb linear argument to
sock_alloc_send_skb if needed for large packets.

The header parsing code does not require an allocated skb, so is
safe to move. Later pass to tpacket_fill_skb the computed data
start and length.

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

packet: vnet_hdr support for tpacket_rcv

Support socket option PACKET_VNET_HDR together with PACKET_RX_RING.
When enabled, a struct virtio_net_hdr will precede the data in the
packet ring slots.

Verified with test program at
github.com/wdebruij/kerneltools/blob/master/tests/psock_rxring_vnet.c

  pkt: 1454269209.798420 len=5066
  vnet: gso_type=tcpv4 gso_size=1448 hlen=66 ecn=off
  csum: start=34 off=16
  eth: proto=0x800
  ip: src=<masked> dst=<masked> proto=6 len=5052

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

packet: move vnet_hdr code to helper functions

packet_snd and packet_rcv support virtio net headers for GSO.
Move this logic into helper functions to be able to reuse it in
tpacket_snd and tpacket_rcv.

This is a straighforward code move with one exception. Instead of
creating and passing a separate gso_type variable, reuse
vnet_hdr.gso_type after conversion from virtio to kernel gso type.

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bonding: 3ad: apply ad_actor settings changes immediately

Currently the bonding allows to set ad_actor_system and prio while the
bond device is down, but these are actually applied only if there aren't
any slaves yet (applied to bond device when first slave shows up, and to
slaves at 3ad bind time). After this patch changes are applied immediately
and the new values can be used/seen after the bond's upped so it's not
necessary anymore to release all and enslave again to see the changes.

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <gospo@cumulusnetworks.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'bridge-mdb-entry-offload-flag'

Jiri Pirko says:

====================
bridge: mdb: flag offloaded mdb entries

This patchset extends uapi to let the user know if an mdb entry is offloaded.
====================

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

bridge: mdb: Passing the port-group pointer to br_mdb module

Passing the port-group to br_mdb in order to allow direct access to the
structure. br_mdb will later use the structure to reflect HW reflection
status via "state" variable.

Signed-off-by: Elad Raz <eladr@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge: mdb: Separate br_mdb_entry->state from net_bridge_port_group->state

Change net_bridge_port_group 'state' member to 'flags' and define new set
of flags internal to the kernel.

Signed-off-by: Elad Raz <eladr@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge: mdb: add support for offloaded mdb entries

Add new bitmask member 'flags' to br_mdb_entry structure. Adding
MDB_FLAGS_OFFLOAD bit which indicates MDB entries is offloaded to hardware.

Signed-off-by: Elad Raz <eladr@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bonding: trivial: style fixes

remove some redudant brackets, use sizeof(*) instead of sizeof(struct x).

Signed-off-by: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: Fix syncookies sysctl default.

Unintentionally the default was changed to zero, fix
that.

Fixes: 12ed8244ed ("ipv4: Namespaceify tcp syncookies sysctl knob")
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'ns-tcp-sysctls'

Nikolay Borisov says:

====================
Namespaceify more of the tcp sysctl knobs

This patch series continues making more of the tcp-related
sysctl knobs be per net-namespace. Most of these apply per
socket and have global defaults so should be safe and I
don't expect any breakages.

Having those per net-namespace is useful when multiple
containers are hosted and it is required to tune the
tcp settings for each independently of the host node.

I've split the patches to be per-sysctl but after
the review if the outcome is positive I'm happy
to either send it in one big blob or just.
====================

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

ipv4: Namespaceify tcp_notsent_lowat sysctl knob

Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Namespaceify tcp_fin_timeout sysctl knob

Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Namespaceify tcp_orphan_retries sysctl knob

Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Namespaceify tcp_retries2 sysctl knob

Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Namespaceify tcp_retries1 sysctl knob

Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Namespaceify tcp reordering sysctl knob

Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Namespaceify tcp syncookies sysctl knob

Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Namespaceify tcp synack retries sysctl knob

Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Namespaceify tcp syn retries sysctl knob

Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'batman-adv-for-davem' of git://git.open-mesh.org/linux-merge

Antonio Quartulli says:

====================
This batch of patches includes a number of corrections and
improvements for our kernel-doc. These changes also make sure
that our doc is now properly processed by the kernel-doc
parsing tool.

Other than that you have a patch updating all the copyright
lines to 2016 and another patch switching the URLs in our
readme, Kconfig and MAINTAINERS file from "http" to "https".
Both by Sven Eckelmann.
====================

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

Merge branch 'virtio_net_ethtool_settings'

Nikolay Aleksandrov says:

====================
virtio_net: add ethtool get/set settings support

Patch 1 adds ethtool speed/duplex validation functions which check if the
value is defined. Patch 2 adds support for ethtool (get|set)_settings and
uses the validation functions to check the user-supplied values.

v2: split in 2 patches to allow everyone to make use of the validation
functions and allow virtio_net devices to be half duplex
v3: added a check to return error if the user tries to change anything else
besides duplex/speed as per Michael's comment
v4: Set port type to PORT_OTHER
v5: clear diff1.port (ignore port) when checking for changes since we set
it now and ethtool uses it in the set request

Sorry about the pointless iterations, should've all covered now.
====================

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

virtio_net: add ethtool support for set and get of settings

This patch allows the user to set and retrieve speed and duplex of the
virtio_net device via ethtool. Having this functionality is very helpful
for simulating different environments and also enables the virtio_net
device to participate in operations where proper speed and duplex are
required (e.g. currently bonding lacp mode requires full duplex). Custom
speed and duplex are not allowed, the user-supplied settings are validated
before applying.

Example:
$ ethtool eth1
Settings for eth1:
...
Speed: Unknown!
Duplex: Unknown! (255)
$ ethtool -s eth1 speed 1000 duplex full
$ ethtool eth1
Settings for eth1:
...
Speed: 1000Mb/s
Duplex: Full

Based on a patch by Roopa Prabhu.

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ethtool: add speed/duplex validation functions

Add functions which check if the speed/duplex are defined.

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'sunvnet-tracepoints'

Sowmini Varadhan says:

====================
sunvnet: perf tracepoint hooks

Added some perf tracepoints to help track and debug sunvnet
descriptor state at run-time.
====================

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

sunvnet: perf tracepoint invocations to trace LDC state machine

Use sunvnet perf trace macros to monitor LDC message exchange state.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sunvnet: Add support for perf LDC event tracing

Add perf event macros for support of tracing and instrumentation
of LDC state machine

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'tcp_cong_ctrl_refactoring'

Yuchung Cheng says:

====================
tcp: congestion control refactoring

This patch set refactors the sequence of congestion control,
loss recovery, and transmission logic in TCP ack processing.

The design goal is to decouple and sequence them in the following order:

  0. ACK accounting: free or tag sent packets [unchanged]

  1. loss recovery: identify lost/ecn packets and update congestion state

  2. congestion control: up/down cwnd and pacing rate based on (1)

  3. transmission: send new or retransmit old based on (1) and (2)

This refactoring makes the cwnd changes more clear because it's done
in one place. The packet accounting is also more robust especially
for connections that do not support SACK. Patch 1-4 and 6 are
refactoring and patch 5 improves TCP performance under reordering.
====================

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

tcp: tcp_cong_control helper

Refactor and consolidate cwnd and rate updates into a new function
tcp_cong_control().

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: make congestion control more robust against reordering

This change enables congestion control to update cwnd based on
not only packet cumulatively acked but also packets delivered
out-of-order. This makes congestion control robust against packet
reordering because it may raise cwnd as long as packets are being
delivered once reordering has been detected (i.e., it only cares
the amount of packets delivered, not the ordering among them).

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: refactor pkts acked accounting

A small refactoring that gets number of packets cumulatively acked
from tcp_clean_rtx_queue() directly.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: new delivery accounting

This patch changes the accounting of how many packets are
newly acked or sacked when the sender receives an ACK.

The current approach basically computes

   newly_acked_sacked = (prior_packets - prior_sacked) -
                        (tp->packets_out - tp->sacked_out)

   where prior_packets and prior_sacked out are snapshot
   at the beginning of the ACK processing.

The new approach tracks the delivery information via a new
TCP state variable "delivered" which monotically increases
as new packets are delivered in order or out-of-order.

The reason for this change is that the current approach is
brittle that produces negative or inaccurate estimate.

   1) For non-SACK connections, an ACK that advances the SND.UNA
   could reset the DUPACK counters (tp->sacked_out) in
   tcp_process_loss() or tcp_fastretrans_alert(). This inflates
   the inflight suddenly and causes under-estimate or even
   negative estimate. Here is a real example:

                   before   after (processing ACK)
   packets_out     75       73
   sacked_out      23        0
   ca state        Loss     Open

   The old approach computes (75-23) - (73 - 0) = -21 delivered
   while the new approach computes 1 delivered since it
   considers the 2nd-24th packets are delivered OOO.

   2) MSS change would re-count packets_out and sacked_out so
   the estimate is in-accurate and can even become negative.
   E.g., the inflight is doubled when MSS is halved.

   3) Spurious retransmission signaled by DSACK is not accounted

The new approach is simpler and more robust. For SACK connections,
tp->delivered increments as packets are being acked or sacked in
SACK and ACK processing.

For non-sack connections, it's done in tcp_remove_reno_sacks() and
tcp_add_reno_sack(). When an ACK advances the SND.UNA, tp->delivered
is incremented by the number of packets ACKed (less the current
number of DUPACKs received plus one packet hole).  Upon receiving
a DUPACK, tp->delivered is incremented assuming one out-of-order
packet is delivered.

Upon receiving a DSACK, tp->delivered is incremtened assuming one
retransmission is delivered in tcp_sacktag_write_queue().

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: move cwnd reduction after recovery state procesing

Currently the cwnd is reduced and increased in various different
places. The reduction happens in various places in the recovery
state processing (tcp_fastretrans_alert) while the increase
happens afterward.

A better sequence is to identify lost packets and update
the congestion control state (icsk_ca_state) first. Then base
on the new state, up/down the cwnd in one central place. It's
more clear to reason cwnd changes.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: retransmit after recovery processing and congestion control

The retransmission and F-RTO transmission currently happen inside
recovery state processing (tcp_fastretrans_alert) but before
congestion control.  This refactoring moves the logic after both
s.t. we can determine how much to send (cwnd) before deciding what to
send.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: drop write-only stack variable

Remove a write-only stack variable from unix_attach_fds(). This is a
left-over from the security fix in:

    commit 712f4aad406bb1ed67f3f98d04c044191f0ff593
    Author: willy tarreau <w@1wt.eu>
    Date:   Sun Jan 10 07:54:56 2016 +0100

        unix: properly account for FDs passed over unix sockets

Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Add support for fill_slave_info to VRF device

Allows userspace to have direct access to VRF table association
versus looking up master device and its table.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

xen-netback: implement dynamic multicast control

My recent patch to the Xen Project documents a protocol for 'dynamic
multicast control' in netif.h. This extends the previous multicast control
protocol to not require a shared ring reconnection to turn the feature off.
Instead the backend watches the "request-multicast-control" key in xenstore
and turns the feature off if the key value is written to zero.

This patch adds support for dynamic multicast control in xen-netback.

Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Cc: Ian Campbell <ian.campbell@citrix.com>
Cc: Wei Liu <wei.liu2@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'be2net-non-critical-fixes'

Sriharsha Basavapatna says:

====================
be2net patch-set

v2 changes:
Patch-4: Changed a tab to space in be.h
Patches-6,7,8: Updated commit log summary line: benet --> be2net

Hi David,

The following patch set contains a few non-critical bug fixes. Please
consider applying this to the net-next tree. Thanks.

Patch-1 fixes be_set_phys_id() ethtool function to return an error code.
Patch-2 fixes a warning when some commands fail for VFs.
Patch-3 fixes be_vlan_rem_vid() to verify vlan being removed is in the list.
Patch-4 improves SRIOV queue distribution logic.
Patch-5 avoids running self test on VFs.
Patch-6 fixes error recovery in Lancer to clean up after moving to ready state.
Patch-7 adds retry logic to error recovery in case of recovery failures
Patch-8 fixes time interval used in eq delay computation routine
====================

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

be2net: Fix interval calculation in interrupt moderation

Interrupt moderation parameters need to be recalculated only
after a time interval of 1 ms. Interval calculation is wrong
when there is a rollover of jiffies. Using recommended way of interval
calculation using jiffies to fix this.

Signed-off-by: Padmanabh Ratnakar <padmanabh.ratnakar@broadcom.com>
Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: Add retry in case of error recovery failure

Retry error recovery MAX_ERR_RECOVERY_RETRY_COUNT times in case of
failure during error recovery.

Signed-off-by: Padmanabh Ratnakar <padmanabh.ratnakar@broadcom.com>
Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: Fix Lancer error recovery

After error is detected, wait for adapter to move to ready state
before destroying queues and cleanup of other resources. Also
skip performing any cleanup for non-Lancer chips and move debug
messages to correct routine.

Signed-off-by: Padmanabh Ratnakar <padmanabh.ratnakar@broadcom.com>
Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: Don't run ethtool self-tests for VFs

The CMD_SUBSYSTEM_LOWLEVEL cmds need DEV_CFG Privilege to run
which VFs don't have by default.
Self-tests need to be issued only for PFs.

Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: SRIOV Queue distribution should factor in EQ-count of VFs

The SRIOV resource distribution logic for RX/TX queue counts is not optimal
when a small number of VFs are enabled. It does not take into account the
VF's EQ count while computing the queue counts. Because of this, the VF
gets a large number of queues, though it doesn't have sufficient EQs,
resulting in wasted queue resources. And the PF gets a smaller share of
queues though it has more EQs. Fix this by capping the VF queue count at
its EQ count.

Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: Fix be_vlan_rem_vid() to check vlan id being removed

The driver decrements its vlan count without checking if it is really
present in its list. This results in an invalid vlan count and impacts
subsequent vlan add/rem ops. The function be_vlan_rem_vid() should be
updated to fix this.

Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: check for INSUFFICIENT_PRIVILEGES error

The driver currently logs the message "VF is not privileged to issue
opcode" by checking only the base_status field for UNAUTHORIZED_REQUEST.
Add check to look for INSUFFICIENT_PRIVILEGES in the additional status
field also as not all cmds fail with that base status.

Signed-off-by: Suresh Reddy <suresh.reddy@broadcom.com>
Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: return error status from be_set_phys_id()

be_set_phys_id() returns 0 to ethtool when the command fails in the FW.

This patch fixes the set_phys_id() to return -EIO in case the FW cmd fails.

Signed-off-by: Suresh Reddy <suresh.reddy@broadcom.com>
Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'vxlan-headers-and-tx-cleanups'

Jiri Benc says:

====================
vxlan: cleanup headers and tx path

This is first part of vxlan cleanup. It makes vxlan.h better organized and
removes code duplication in the tx path. There's no change in functionality.

This is in preparation for VXLAN-GPE support. Other sets will follow with
cleanup of rx path, modifications of vxlan interface setup and
cleanup/modification of fdb handling. The goal of these patchsets is to
consolidate VXLAN extension handling (right now it's scattered all around
the code) and allow vxlan to operate in L3 mode (i.e. without inner Ethernet
headers).

The full picture (all 30 patches) can be seen at:
https://github.com/jbenc/linux-vxlan/commits/master

Changelog:
v2: added patch 5/6, fixing the udp sum problem spotted by Paolo Abeni
====================

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

vxlan: consolidate vxlan_xmit_skb and vxlan6_xmit_skb

There's a lot of code duplication. Factor out the duplicate code to a new
function shared between IPv4 and IPv6 xmit path.

Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

vxlan: consolidate csum flag handling

The flag for tx checksumming for tunneling over IPv4 and IPv6 is different.
Decide whether to do tx checksumming in vxlan_xmit_one and pass it on as
a separate flag. This will allow for tx path consolidation in the next
patch.

Unfortunately, gcc is not clever enough to see that udp_sum is always
initialized and gives an uninitialized variable warning. Set it to false to
silence the warning.

Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

vxlan: consolidate output route calculation

The code for output route lookup is duplicated for ndo_start_xmit and
ndo_fill_metadata_dst. Move it to a common function.

Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

vxlan: restructure vxlan.h definitions

RCO and GBP are VXLAN extensions, not specified in RFC 7348. Because of
that, they need to be explicitly enabled when creating vxlan interface. By
default, those extensions are not used and plain VXLAN header is sent and
received.

Reflect this in vxlan.h: first, the plain VXLAN header is defined. Following
it, RCO is documented and defined, and likewise for GBP.

Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

vxlan: remove duplicated macros

VNI_HASH_BITS and VNI_HASH_SIZE are defined twice. Remove the extra
definitions.

Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

vxlan: cleanup types

include/net/vxlan.h is a kernel header, no need to prefix fixed size types
with double underscore.

Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: fastopen: call tcp_fin() if FIN present in SYNACK

When we acknowledge a FIN, it is not enough to ack the sequence number
and queue the skb into receive queue. We also have to call tcp_fin()
to properly update socket state and send proper poll() notifications.

It seems we also had the problem if we received a SYN packet with the
FIN flag set, but it does not seem an urgent issue, as no known
implementation can do that.

Fixes: 61d2bcae99f6 ("tcp: fastopen: accept data/FIN present in SYNACK message")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'tipc-topology-updates'

Parthasarathy Bhuvaragan says:

====================
tipc: cleanups, fixes & improvements for topology server

This series contains topology server cleanups, fixes and improvements.

Cleanups in #1-#4:
We remove duplicate data structures and aligin the rest of the code accordingly.

Fixes in #5-#8:
The bugs occur either during configuration or while running on SMP targets,
which are race conditions that pop up under different situations.

Improvements in #9-#10:
Updates to decrease timer usage and improve readability.

v2: Updated commit message in patch 6 based on feedback from
    Sergei Shtylyov sergei.shtylyov@cogentembedded.com
====================

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

tipc: use alloc_ordered_workqueue() instead of WQ_UNBOUND w/ max_active = 1

Until now, tipc_rcv and tipc_send workqueues in server are allocated
with parameters WQ_UNBOUND & max_active = 1.
This parameters passed to this function makes it equivalent to
alloc_ordered_workqueue(). The later form is more explicit and
can inherit future ordered_workqueue changes.

In this commit we replace alloc_workqueue() with more readable
alloc_ordered_workqueue().

Acked-by: Ying Xue <ying.xue@windriver.com>
Reviewed-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>

tipc: donot create timers if subscription timeout = TIPC_WAIT_FOREVER

Until now, we create timers even for the subscription requests
with timeout = TIPC_WAIT_FOREVER.
This can be improved by avoiding timer creation when the timeout
is set to TIPC_WAIT_FOREVER.

In this commit, we introduce a check to creates timers only
when timeout != TIPC_WAIT_FOREVER.

Acked-by: Ying Xue <ying.xue@windriver.com>
Reviewed-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>

tipc: protect tipc_subscrb_get() with subscriber spin lock

Until now, during subscription creation the mod_time() &
tipc_subscrb_get() are called after releasing the subscriber
spin lock.

In a SMP system when performing a subscription creation, if the
subscription timeout occurs simultaneously (the timer is
scheduled to run on another CPU) then the timer thread
might decrement the subscribers refcount before the create
thread increments the refcount.

This can be simulated by creating subscription with timeout=0 and
sometimes the timeout occurs before the create request is complete.
This leads to the following message:
[30.702949] BUG: spinlock bad magic on CPU#1, kworker/u8:3/87
[30.703834] general protection fault: 0000 [#1] SMP
[30.704826] CPU: 1 PID: 87 Comm: kworker/u8:3 Not tainted 4.4.0-rc8+ #18
[30.704826] Workqueue: tipc_rcv tipc_recv_work [tipc]
[30.704826] task: ffff88003f878600 ti: ffff88003fae0000 task.ti: ffff88003fae0000
[30.704826] RIP: 0010:[<ffffffff8109196c>]  [<ffffffff8109196c>] spin_dump+0x5c/0xe0
[...]
[30.704826] Call Trace:
[30.704826]  [<ffffffff81091a16>] spin_bug+0x26/0x30
[30.704826]  [<ffffffff81091b75>] do_raw_spin_lock+0xe5/0x120
[30.704826]  [<ffffffff81684439>] _raw_spin_lock_bh+0x19/0x20
[30.704826]  [<ffffffffa0096f10>] tipc_subscrb_rcv_cb+0x1d0/0x330 [tipc]
[30.704826]  [<ffffffffa00a37b1>] tipc_receive_from_sock+0xc1/0x150 [tipc]
[30.704826]  [<ffffffffa00a31df>] tipc_recv_work+0x3f/0x80 [tipc]
[30.704826]  [<ffffffff8106a739>] process_one_work+0x149/0x3c0
[30.704826]  [<ffffffff8106aa16>] worker_thread+0x66/0x460
[30.704826]  [<ffffffff8106a9b0>] ? process_one_work+0x3c0/0x3c0
[30.704826]  [<ffffffff8106a9b0>] ? process_one_work+0x3c0/0x3c0
[30.704826]  [<ffffffff8107029d>] kthread+0xed/0x110
[30.704826]  [<ffffffff810701b0>] ? kthread_create_on_node+0x190/0x190
[30.704826]  [<ffffffff81684bdf>] ret_from_fork+0x3f/0x70

In this commit,
1. we remove the check for the return code for mod_timer()
2. we protect tipc_subscrb_get() using the subscriber spin lock.
   We increment the subscriber's refcount as soon as we add the
   subscription to subscriber's subscription list.

Acked-by: Ying Xue <ying.xue@windriver.com>
Reviewed-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>

tipc: hold subscriber->lock for tipc_nametbl_subscribe()

Until now, while creating a subscription the subscriber lock
protects only the subscribers subscription list and not the
nametable. The call to tipc_nametbl_subscribe() is outside
the lock. However, at subscription timeout and cancel both
the subscribers subscription list and the nametable are
protected by the subscriber lock.

This asymmetric locking mechanism leads to the following problem:
In a SMP system, the timer can be fire on another core before
the create request is complete.
When the timer thread calls tipc_nametbl_unsubscribe() before create
thread calls tipc_nametbl_subscribe(), we get a nullptr exception.

This can be simulated by creating subscription with timeout=0 and
sometimes the timeout occurs before the create request is complete.

The following is the oops:
[57.569661] BUG: unable to handle kernel NULL pointer dereference at (null)
[57.577498] IP: [<ffffffffa02135aa>] tipc_nametbl_unsubscribe+0x8a/0x120 [tipc]
[57.584820] PGD 0
[57.586834] Oops: 0002 [#1] SMP
[57.685506] CPU: 14 PID: 10077 Comm: kworker/u40:1 Tainted: P OENX 3.12.48-52.27.1.     9688.1.PTF-default #1
[57.703637] Workqueue: tipc_rcv tipc_recv_work [tipc]
[57.708697] task: ffff88064c7f00c0 ti: ffff880629ef4000 task.ti: ffff880629ef4000
[57.716181] RIP: 0010:[<ffffffffa02135aa>]  [<ffffffffa02135aa>] tipc_nametbl_unsubscribe+0x8a/   0x120 [tipc]
[...]
[57.812327] Call Trace:
[57.814806]  [<ffffffffa0211c77>] tipc_subscrp_delete+0x37/0x90 [tipc]
[57.821357]  [<ffffffffa0211e2f>] tipc_subscrp_timeout+0x3f/0x70 [tipc]
[57.827982]  [<ffffffff810618c1>] call_timer_fn+0x31/0x100
[57.833490]  [<ffffffff81062709>] run_timer_softirq+0x1f9/0x2b0
[57.839414]  [<ffffffff8105a795>] __do_softirq+0xe5/0x230
[57.844827]  [<ffffffff81520d1c>] call_softirq+0x1c/0x30
[57.850150]  [<ffffffff81004665>] do_softirq+0x55/0x90
[57.855285]  [<ffffffff8105aa35>] irq_exit+0x95/0xa0
[57.860290]  [<ffffffff815215b5>] smp_apic_timer_interrupt+0x45/0x60
[57.866644]  [<ffffffff8152005d>] apic_timer_interrupt+0x6d/0x80
[57.872686]  [<ffffffffa02121c5>] tipc_subscrb_rcv_cb+0x2a5/0x3f0 [tipc]
[57.879425]  [<ffffffffa021c65f>] tipc_receive_from_sock+0x9f/0x100 [tipc]
[57.886324]  [<ffffffffa021c826>] tipc_recv_work+0x26/0x60 [tipc]
[57.892463]  [<ffffffff8106fb22>] process_one_work+0x172/0x420
[57.898309]  [<ffffffff8107079a>] worker_thread+0x11a/0x3c0
[57.903871]  [<ffffffff81077114>] kthread+0xb4/0xc0
[57.908751]  [<ffffffff8151f318>] ret_from_fork+0x58/0x90

In this commit, we do the following at subscription creation:
1. set the subscription's subscriber pointer before performing
   tipc_nametbl_subscribe(), as this value is required further in
   the call chain ex: by tipc_subscrp_send_event().
2. move tipc_nametbl_subscribe() under the scope of subscriber lock

Acked-by: Ying Xue <ying.xue@windriver.com>
Reviewed-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>

tipc: fix connection abort when receiving invalid cancel request

Until now, the subscribers endianness for a subscription
create/cancel request is determined as:
    swap = !(s->filter & (TIPC_SUB_PORTS | TIPC_SUB_SERVICE))
The checks are performed only for port/service subscriptions.

The swap calculation is incorrect if the filter in the subscription
cancellation request is set to TIPC_SUB_CANCEL (it's a malformed
cancel request, as the corresponding subscription create filter
is missing).
Thus, the check if the request is for cancellation fails and the
request is treated as a subscription create request. The
subscription creation fails as the request is illegal, which
terminates this connection.

In this commit we determine the endianness by including
TIPC_SUB_CANCEL, which will set swap correctly and the
request is processed as a cancellation request.

Acked-by: Ying Xue <ying.xue@windriver.com>
Reviewed-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>

tipc: fix connection abort during subscription cancellation

In 'commit 7fe8097cef5f ("tipc: fix nullpointer bug when subscribing
to events")', we terminate the connection if the subscription
creation fails.
In the same commit, the subscription creation result was based on
the value of subscription pointer (set in the function) instead of
the return code.

Unfortunately, the same function also handles subscription
cancellation request. For a subscription cancellation request,
the subscription pointer cannot be set. Thus the connection is
terminated during cancellation request.

In this commit, we move the subcription cancel check outside
of tipc_subscrp_create(). Hence,
- tipc_subscrp_create() will create a subscripton
- tipc_subscrb_rcv_cb() will subscribe or cancel a subscription.

Fixes: 'commit 7fe8097cef5f ("tipc: fix nullpointer bug when subscribing to events")'
Acked-by: Ying Xue <ying.xue@windriver.com>
Reviewed-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>

tipc: introduce tipc_subscrb_subscribe() routine

In this commit, we split tipc_subscrp_create() into two:
1. tipc_subscrp_create() creates a subscription
2. A new function tipc_subscrp_subscribe() adds the
   subscription to the subscriber subscription list,
   activates the subscription timer and subscribes to
   the nametable updates.

In future commits, the purpose of tipc_subscrb_rcv_cb() will
be to either subscribe or cancel a subscription.

There is no functional change in this commit.

Acked-by: Ying Xue <ying.xue@windriver.com>
Reviewed-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>

tipc: remove struct tipc_name_seq from struct tipc_subscription

Until now, struct tipc_subscriber has duplicate fields for
type, upper and lower (as member of struct tipc_name_seq) at:
1. as member seq in struct tipc_subscription
2. as member seq in struct tipc_subscr, which is contained
   in struct tipc_event
The former structure contains the type, upper and lower
values in network byte order and the later contains the
intact copy of the request.
The struct tipc_subscription contains a field swap to
determine if request needs network byte order conversion.
Thus by using swap, we can convert the request when
required instead of duplicating it.

In this commit,
1. we remove the references to these elements as members of
   struct tipc_subscription and replace them with elements
   from struct tipc_subscr.
2. provide new functions to convert the user request into
   network byte order.

Acked-by: Ying Xue <ying.xue@windriver.com>
Reviewed-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>