David Ahern says:

====================
net: VRF support in IPv6 stack

Initial support for VRF in IPv6 stack. Makes IPv6 functionality on par
with IPv4 -- ping, tcp client/server and udp client/server all work fine.
tcpdump on vrf device and external tap (e.g., host side tap device) shows
all packets with proper addresses. IPv6 does not need the source address
operation like IPv4. Verified vti6 works properly in my setup as does use
of an IPv6 address on the VRF device.

v3
- re-based to top of net-next (updates per net namespace changes by Eric)
- fixed dst_entry typecasts as requested by Dave
- added flags to inet6_rtm_getroute (IPv6 version of deaa0a6a)

v2
- fixed CONFIG_IPV6 dependency as questioned by Cong
  - if IPV6 is a module, kbuild ensures VRF is a module
  - if IPV6 is disabled IPV6 functionality is compiled out of VRF module
- addressed comments from Nik over IRC
  - removed duplicate call to netif_is_l3_master in l3mdev_rt6_dst_by_oif
  - changed allocation flag from GFP_ATOMIC to GFP_KERNEL since it is init time
  - added free of rt6i_pcpu
  - check_ipv6_frame returns false only if packet is NDISC type
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

As with IPv4 support for VRFs added to IPv6 stack by replacing hardcoded
table ids with possibly device specific ones and manipulating the oif in
the flowi6. The flow flags are used to skip oif compare in nexthop lookups
if the device is enslaved to a VRF via the L3 master device.
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add support for IPv6 to VRF device driver. Implemenation parallels what
has been done for IPv4.
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

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

Add operations to retrieve cached IPv6 dst entry from l3mdev device
and lookup IPv6 source address.
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

commit c62987bb ("bridge: push bridge setting ageing_time down to
switchdev") introduced a timer race condition because the gc_timer can
get rearmed after it's supposedly stopped and flushed in br_dev_delete()
leading to a use of freed memory. So take rtnl to sync with bridge
destruction when setting ageing_timer.
Here's the trace reproduced with these two commands running in parallel:
while :; do echo 10000 > /sys/class/net/br0/bridge/ageing_timer; done;
while :; do brctl addbr br0; ip l set br0 up; ip l set br0 down;
brctl delbr br0; done;

[  300.000029] BUG: unable to handle kernel paging request at
ffffffff811c59d3
[  300.000263] IP: [<ffffffff810f168e>] __internal_add_timer+0x2e/0xd0
[  300.000422] PGD 1a0f067 PUD 1a10063 PMD 10001e1
[  300.000639] Oops: 0003 [#1] SMP
[  300.000793] Modules linked in: bridge stp llc nfsd auth_rpcgss
oid_registry nfs_acl nfs lockd grace fscache sunrpc crct10dif_pclmul
crc32_pclmul crc32c_intel ghash_clmulni_intel ppdev aesni_intel
aes_x86_64 glue_helper lrw gf128mul ablk_helper cryptd
snd_hda_codec_generic qxl drm_kms_helper psmouse pcspkr ttm
snd_hda_intel 9pnet_virtio evdev serio_raw joydev snd_hda_codec 9pnet
virtio_balloon drm snd_hwdep virtio_console snd_hda_core pvpanic snd_pcm
i2c_piix4 snd_timer acpi_cpufreq parport_pc snd parport soundcore button
processor i2c_core ipv6 autofs4 hid_generic usbhid hid ext4 crc16
mbcache jbd2 sg sr_mod cdrom ata_generic virtio_blk virtio_net e1000
ehci_pci uhci_hcd ehci_hcd usbcore usb_common floppy ata_piix libata
virtio_pci virtio_ring virtio scsi_mod
[  300.004008] CPU: 1 PID: 1169 Comm: bash Not tainted 4.3.0-rc3+ #46
[  300.004008] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
[  300.004008] task: ffff880035be2200 ti: ffff88003795c000 task.ti:
ffff88003795c000
[  300.004008] RIP: 0010:[<ffffffff810f168e>]  [<ffffffff810f168e>]
__internal_add_timer+0x2e/0xd0
[  300.004008] RSP: 0018:ffff88003fd03e78  EFLAGS: 00010046
[  300.004008] RAX: ffff88003fd0ef60 RBX: 840fc78949c08548 RCX:
00000001ffffffff
[  300.004008] RDX: 0000000000000000 RSI: ffffffff811c59d3 RDI:
ffff88003fd0df00
[  300.004008] RBP: ffff88003fd03e78 R08: 00000000ffffffff R09:
0000000000000000
[  300.004008] R10: 0000000000000000 R11: 0000000000000000 R12:
ffff88003fd0df00
[  300.004008] R13: 0000000000000000 R14: 0000000000000001 R15:
ffffffff816032e0
[  300.004008] FS:  00007fcbdd609700(0000) GS:ffff88003fd00000(0000)
knlGS:0000000000000000
[  300.004008] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  300.004008] CR2: ffffffff811c59d3 CR3: 0000000037879000 CR4:
00000000000406e0
[  300.004008] Stack:
[  300.004008]  ffff88003fd03ea8 ffffffff810f1775 ffff88003c8cb958
ffff88003fd0df00
[  300.004008]  0000000000000000 0000000000000001 ffff88003fd03f18
ffffffff810f28c4
[  300.004008]  ffff88003fd0eb68 ffff88003fd0e968 ffff88003fd0e768
ffff88003fd0df68
[  300.004008] Call Trace:
[  300.004008]  <IRQ>
[  300.004008]  [<ffffffff810f1775>] cascade+0x45/0x70
[  300.004008]  [<ffffffff810f28c4>] run_timer_softirq+0x2f4/0x340
[  300.004008]  [<ffffffff8107e380>] __do_softirq+0xd0/0x440
[  300.004008]  [<ffffffff8107e8a3>] irq_exit+0xb3/0xc0
[  300.004008]  [<ffffffff815c2032>] smp_apic_timer_interrupt+0x42/0x50
[  300.004008]  [<ffffffff815bfe37>] apic_timer_interrupt+0x87/0x90
[  300.004008]  <EOI>
[  300.004008]  [<ffffffff811fb80c>] ? create_object+0x13c/0x2e0
[  300.004008]  [<ffffffff8109b23e>] ? __kernel_text_address+0x4e/0x70
[  300.004008]  [<ffffffff8109b23e>] ? __kernel_text_address+0x4e/0x70
[  300.004008]  [<ffffffff8101e17f>] print_context_stack+0x7f/0xf0
[  300.004008]  [<ffffffff8101d55b>] dump_trace+0x11b/0x300
[  300.004008]  [<ffffffff8102970b>] save_stack_trace+0x2b/0x50
[  300.004008]  [<ffffffff811fb80c>] create_object+0x13c/0x2e0
[  300.004008]  [<ffffffff815b2e8e>] kmemleak_alloc+0x4e/0xb0
[  300.004008]  [<ffffffff811e475d>] kmem_cache_alloc_trace+0x18d/0x2f0
[  300.004008]  [<ffffffff8128b139>] kernfs_fop_open+0xc9/0x380
[  300.004008]  [<ffffffff8120214f>] do_dentry_open+0x1ff/0x2f0
[  300.004008]  [<ffffffff8128b070>] ? kernfs_fop_release+0x70/0x70
[  300.004008]  [<ffffffff812034f9>] vfs_open+0x59/0x60
[  300.004008]  [<ffffffff812130de>] path_openat+0x1ce/0x1260
[  300.004008]  [<ffffffff812154ae>] do_filp_open+0x7e/0xe0
[  300.004008]  [<ffffffff812251ff>] ? __alloc_fd+0xaf/0x180
[  300.004008]  [<ffffffff8120387b>] do_sys_open+0x12b/0x210
[  300.004008]  [<ffffffff8120397e>] SyS_open+0x1e/0x20
[  300.004008]  [<ffffffff815bf0b6>] entry_SYSCALL_64_fastpath+0x16/0x7a
[  300.004008] Code: 66 90 48 8b 46 10 48 8b 4f 40 55 48 89 c2 48 89 e5
48 29 ca 48 81 fa ff 00 00 00 77 20 0f b6 c0 48 8d 44 c7 68 48 8b 10 48
85 d2 <48> 89 16 74 04 48 89 72 08 48 89 30 48 89 46 08 5d c3 48 81 fa
[  300.004008] RIP  [<ffffffff810f168e>] __internal_add_timer+0x2e/0xd0
[  300.004008]  RSP <ffff88003fd03e78>
[  300.004008] CR2: ffffffff811c59d3

Fixes: c62987bb ("bridge: push bridge setting ageing_time down to switchdev")
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

We shouldn't allow BRIDGE_VLAN_INFO_PVID flag in VLAN ranges.
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Acked-by: Elad Raz <eladr@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Vivien Didelot says:

====================
net: dsa: mv88e6xxx: fix hardware bridging

DSA and its drivers currently hook the NETDEV_CHANGEUPPER net_device event in
order to configure the VLAN map of every port.

This VLAN map is a feature of these switch chips to hardcode and restrict which
output ports a given input port can egress frames to.

A Linux bridge is a simple untagged VLAN propagated by the bridge code itself.
With a proper 802.1Q support, a driver does not need this hook anymore, and
will simply program the related VLAN object.

This patchset improves the hardware bridging code in the mv88e6xxx driver with
a strict 802.1Q mode.

Ideally, the equivalent must be done for Broadcom Starfighter 2 and Rocker,
before completely getting rid of this hook.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Playing with the VLAN map of every port to implement "hardware bridging"
in the 88E6352 driver was a hack until full 802.1Q was supported.

Indeed with 802.1Q port mode "Disabled" or "Fallback", this feature is
used to restrict which output ports an input port can egress frames to.

A Linux bridge is an untagged VLAN. With full 802.1Q support, we don't
need this hack anymore and can use the "Secure" strict 802.1Q port mode.

With this mode, the port-based VLAN map still needs to be configured,
but all the logic is VTU-centric. This means that the switch only cares
about rules described in its hardware VLAN table, which is exactly what
Linux bridge expects and what we want.

Note also that the hardware bridging was broken with the previous
flexible "Fallback" 802.1Q port mode. Here's an example:

Port0 and Port1 belong to the same bridge. If Port0 sends crafted tagged
frames with VID 200 to Port1, Port1 receives it. Even if Port1 is in
hardware VLAN 200, but not Port0, Port1 will still receive it, because
Fallback mode doesn't care about invalid VID or non-member source port.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

A DSA driver may not provide the port_join_bridge and port_leave_bridge
functions, so don't warn in such case.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Since we configure a switch chip through a Linux bridge, and a bridge is
implemented as a VLAN, there is no need for per-port FID anymore.

This patch gets rid of this and simplifies the driver code since we can
now directly map all 4095 FIDs available to all VLANs.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

With 88E6352 and similar switch chips, each port has a map to restrict
which output port this input port can egress frames to.

The current driver code implements hardware bridging using this feature,
and assigns to a bridge group the FID of its first member.

Now that 802.1Q is fully implemented in this driver, a Linux bridge
which is a simple untagged VLAN, already gets its own FID.

This patch gets rid of the per-bridge FID and explicits the usage of the
port based VLAN map feature.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Consider the following "duelling syn" sequence between two peers A and B:
        	A		B
        	SYN1     -->
        	    	<--	SYN2
        	SYN2ACK  -->

Note that the SYN/ACK has already been sent out by TCP before
rds_tcp_accept_one() gets invoked as part of callbacks.

If the inet_addr(A) is numerically less than inet_addr(B),
the arbitration scheme in rds_tcp_accept_one() will prefer the
TCP connection triggered by SYN1, and will send a CLOSE for the
SYN2 (just after the SYN2ACK was sent).

Since B also follows the same arbitration scheme, it will send the SYN-ACK
for SYN1 that will set up a healthy ESTABLISHED connection on both sides.
B will also get a  CLOSE for SYN2, which should result in the cleanup
of the TCP state machine for SYN2, but it should not trigger any
stale RDS-TCP callbacks (such as ->writespace, ->state_change etc),
that would disrupt the progress of the SYN2 based RDS-TCP  connection.

Thus the arbitration scheme in rds_tcp_accept_one() should restore
rds_tcp callbacks for the winner before setting them up for the
new accept socket, and also make sure that conn->c_outgoing
is set to 0 so that we do not trigger any reconnect attempts on the
passive side of the tcp socket in the future, in conformance with
commit c82ac7e6 ("net/rds: RDS-TCP: only initiate reconnect attempt
on outgoing TCP socket.")
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The IP address passed to rds_bind() should be vetted by the
transport's ->laddr_check() for a previously bound transport.
This needs to be done to avoid cases where, for example,
the application has asked for an IB transport,
but the IP address passed to bind is only usable on
ethernet interfaces.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The only instance of a qlcnic_mbx_ops structure is never modified.  Thus
the declaration of the structure and all references to the structure type
can be made const.

In the definition of the qlcnic_mailbox structure, the ops field is no
longer lined up with the other fields.  This was left as is, to avoid a lot
of trivial changes on the other lines.

Done with the help of Coccinelle.
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Acked-by: Sony Chacko <sony.chacko@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently it's possible for someone to send a vlan range to the kernel
with the pvid flag set which will result in the pvid bouncing from a
vlan to vlan and isn't correct, it also introduces problems for hardware
where it doesn't make sense having more than 1 pvid. iproute2 already
enforces this, so let's enforce it on kernel-side as well.
Reported-by: Elad Raz <eladr@mellanox.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fix a smatch warning:
drivers/atm/iphase.c:1178 rx_pkt() warn: curly braces intended?

The code is correct, the indention is misleading. In case the allocation
of skb fails, we want to skip to the end.
Signed-off-by: Tillmann Heidsieck <theidsieck@leenox.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Smatch complains about returning hard coded error codes, silence this
warning.

drivers/atm/iphase.c:115 ia_enque_rtn_q() warn: returning -1 instead of -ENOMEM is sloppy
Signed-off-by: Tillmann Heidsieck <theidsieck@leenox.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch makes ip6_route_info_create return err pointer instead of
returning the rt pointer by reference as suggested  by Dave
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch fix the building error reported by Jiri Pirko <jiri@resnulli.us>

drivers/net/ethernet/hisilicon/hns/hnae.h:465:2: error: unknown type
name 'phy_interface_t'
        phy_interface_t phy_if;
	^
the full build log is on https://lists.01.org/pipermail/kbuild-all.
Signed-off-by: huangdaode <huangdaode@hisilicon.com>
Signed-off-by: yankejian <yankejian@huawei.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

timewait or request sockets are small and do not contain sk->sk_tsflags

Without this fix, we might read garbage, and crash later in

__skb_complete_tx_timestamp()
 -> sock_queue_err_skb()

(These pseudo sockets do not have an error queue either)

Fixes: ca6fb065 ("tcp: attach SYNACK messages to request sockets instead of listener")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Eric W. Biederman says:

====================
net: Pass net into defragmentation

This is the next installment of my work to pass struct net through the
output path so the code does not need to guess how to figure out which
network namespace it is in, and ultimately routes can have output
devices in another network namespace.

In netfilter and af_packet we defragment packets in the output path,
and there is the usual amount of confusion about how to compute which
net we are processing the packets in.  This patchset clears that
confusion up by explicitly passing in struct net in ip_defrag,
ip_check_defrag, and nf_ct_frag6_gather.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The function nf_ct_frag6_gather is called on both the input and the
output paths of the networking stack.  In particular ipv6_defrag which
calls nf_ct_frag6_gather is called from both the the PRE_ROUTING chain
on input and the LOCAL_OUT chain on output.

The addition of a net parameter makes it explicit which network
namespace the packets are being reassembled in, and removes the need
for nf_ct_frag6_gather to guess.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

The function ip_defrag is called on both the input and the output
paths of the networking stack.  In particular conntrack when it is
tracking outbound packets from the local machine calls ip_defrag.

So add a struct net parameter and stop making ip_defrag guess which
network namespace it needs to defragment packets in.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip_call_ra_chain is called early in the forwarding chain from
ip_forward and ip_mr_input, which makes skb->dev the correct
expression to get the input network device and dev_net(skb->dev) a
correct expression for the network namespace the packet is being
processed in.

Compute the network namespace and store it in a variable to make the
code clearer.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Recent TCP listener patches exposed a prior af_packet bug :
match_fanout_group() blindly assumes it is always safe
to cast sk to a packet socket to compare fanout with af_packet_priv

But SYNACK packets can be sent while attached to request_sock, which
are smaller than a "struct sock".

We can read non existent memory and crash.

Fixes: c0de08d0 ("af_packet: don't emit packet on orig fanout group")
Fixes: ca6fb065 ("tcp: attach SYNACK messages to request sockets instead of listener")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Eric Leblond <eric@regit.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'wireless-drivers-next-for-davem-2015-10-09' of git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

====================
Major changes:

iwlwifi

* some debugfs improvements
* fix signedness in beacon statistics
* deinline some functions to reduce size when device tracing is enabled
* filter beacons out in AP mode when no stations are associated
* deprecate firmwares version -12
* fix a runtime PM vs. legacy suspend race
* one-liner fix for a ToF bug
* clean-ups in the rx code
* small debugging improvement
* fix WoWLAN with new firmware versions
* more clean-ups towards multiple RX queues;
* some rate scaling fixes and improvements;
* some time-of-flight fixes;
* other generic improvements and clean-ups;

brcmfmac

* rework code dealing with multiple interfaces
* allow logging firmware console using debug level
* support for BCM4350, BCM4365, and BCM4366 PCIE devices
* fixed for legacy P2P and P2P device handling
* correct set and get tx-power

ath9k

* add support for Outside Context of a BSS (OCB) mode

mwifiex

* add USB multichannel feature
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch allows configuring how the source address of ICMP
redirect messages is selected; by default the old behaviour is
retained, while setting icmp_redirects_use_orig_daddr force the
usage of the destination address of the packet that caused the
redirect.

The new behaviour fits closely the RFC 5798 section 8.1.1, and fix the
following scenario:

Two machines are set up with VRRP to act as routers out of a subnet,
they have IPs x.x.x.1/24 and x.x.x.2/24, with VRRP holding on to
x.x.x.254/24.

If a host in said subnet needs to get an ICMP redirect from the VRRP
router, i.e. to reach a destination behind a different gateway, the
source IP in the ICMP redirect is chosen as the primary IP on the
interface that the packet arrived at, i.e. x.x.x.1 or x.x.x.2.

The host will then ignore said redirect, due to RFC 1122 section 3.2.2.2,
and will continue to use the wrong next-op.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Some drivers need to implement both switchdev vlan ops and
vid_add/kill ndos. For that to work in bridge code, we need to try
switchdev op first when adding/deleting vlan id.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In bnx2_init_board, missing free temp_stats_blk on error path when
some operations do failed. Just add the 'kfree' operation.
Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Eric Dumazet says:

====================
tcp: better smp listener behavior

As promised in last patch series, we implement a better SO_REUSEPORT
strategy, based on cpu hints if given by the application.

We also moved sk_refcnt out of the cache line containing the lookup
keys, as it was considerably slowing down smp operations because
of false sharing. This was simpler than converting listen sockets
to conventional RCU (to avoid sk_refcnt dirtying)

Could process 6.0 Mpps SYN instead of 4.2 Mpps on my test server.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Reducing tcp_timewait_sock from 280 bytes to 272 bytes
allows SLAB to pack 15 objects per page instead of 14 (on x86)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

One 32bit hole is following skc_refcnt, use it.
skc_incoming_cpu can also be an union for request_sock rcv_wnd.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sk->sk_refcnt is dirtied for every TCP/UDP incoming packet.
This is a performance issue if multiple cpus hit a common socket,
or multiple sockets are chained due to SO_REUSEPORT.

By moving sk_refcnt 8 bytes further, first 128 bytes of sockets
are mostly read. As they contain the lookup keys, this has
a considerable performance impact, as cpus can cache them.

These 8 bytes are not wasted, we use them as a place holder
for various fields, depending on the socket type.

Tested:
 SYN flood hitting a 16 RX queues NIC.
 TCP listener using 16 sockets and SO_REUSEPORT
 and SO_INCOMING_CPU for proper siloing.

 Could process 6.0 Mpps SYN instead of 4.2 Mpps

 Kernel profile looked like :
    11.68%  [kernel]  [k] sha_transform
     6.51%  [kernel]  [k] __inet_lookup_listener
     5.07%  [kernel]  [k] __inet_lookup_established
     4.15%  [kernel]  [k] memcpy_erms
     3.46%  [kernel]  [k] ipt_do_table
     2.74%  [kernel]  [k] fib_table_lookup
     2.54%  [kernel]  [k] tcp_make_synack
     2.34%  [kernel]  [k] tcp_conn_request
     2.05%  [kernel]  [k] __netif_receive_skb_core
     2.03%  [kernel]  [k] kmem_cache_alloc
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

SO_INCOMING_CPU as added in commit 2c8c56e1 was a getsockopt() command
to fetch incoming cpu handling a particular TCP flow after accept()

This commits adds setsockopt() support and extends SO_REUSEPORT selection
logic : If a TCP listener or UDP socket has this option set, a packet is
delivered to this socket only if CPU handling the packet matches the specified
one.

This allows to build very efficient TCP servers, using one listener per
RX queue, as the associated TCP listener should only accept flows handled
in softirq by the same cpu.
This provides optimal NUMA behavior and keep cpu caches hot.

Note that __inet_lookup_listener() still has to iterate over the list of
all listeners. Following patch puts sk_refcnt in a different cache line
to let this iteration hit only shared and read mostly cache lines.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Edward Hyunkoo Jee <edjee@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

It's useful to allow users to set fwmark for an individual packet,
without changing the socket state. The function this patch adds in
sock layer can be used by the protocols that need such a feature.
Signed-off-by: Edward Hyunkoo Jee <edjee@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Alexei Starovoitov says:

====================
bpf: unprivileged

v1-v2:
- this set logically depends on cb patch
  "bpf: fix cb access in socket filter programs":
  http://patchwork.ozlabs.org/patch/527391/
  which is must have to allow unprivileged programs.
  Thanks Daniel for finding that issue.
- refactored sysctl to be similar to 'modules_disabled'
- dropped bpf_trace_printk
- split tests into separate patch and added more tests
  based on discussion

v1 cover letter:
I think it is time to liberate eBPF from CAP_SYS_ADMIN.
As was discussed when eBPF was first introduced two years ago
the only piece missing in eBPF verifier is 'pointer leak detection'
to make it available to non-root users.
Patch 1 adds this pointer analysis.
The eBPF programs, obviously, need to see and operate on kernel addresses,
but with these extra checks they won't be able to pass these addresses
to user space.
Patch 2 adds accounting of kernel memory used by programs and maps.
It changes behavoir for existing root users, but I think it needs
to be done consistently for both root and non-root, since today
programs and maps are only limited by number of open FDs (RLIMIT_NOFILE).
Patch 2 accounts program's and map's kernel memory as RLIMIT_MEMLOCK.

Unprivileged eBPF is only meaningful for 'socket filter'-like programs.
eBPF programs for tracing and TC classifiers/actions will stay root only.

In parallel the bpf fuzzing effort is ongoing and so far
we've found only one verifier bug and that was already fixed.
The 'constant blinding' pass also being worked on.
It will obfuscate constant-like values that are part of eBPF ISA
to make jit spraying attacks even harder.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Add new tests samples/bpf/test_verifier:

unpriv: return pointer
  checks that pointer cannot be returned from the eBPF program

unpriv: add const to pointer
unpriv: add pointer to pointer
unpriv: neg pointer
  checks that pointer arithmetic is disallowed

unpriv: cmp pointer with const
unpriv: cmp pointer with pointer
  checks that comparison of pointers is disallowed
  Only one case allowed 'void *value = bpf_map_lookup_elem(..); if (value == 0) ...'

unpriv: check that printk is disallowed
  since bpf_trace_printk is not available to unprivileged

unpriv: pass pointer to helper function
  checks that pointers cannot be passed to functions that expect integers
  If function expects a pointer the verifier allows only that type of pointer.
  Like 1st argument of bpf_map_lookup_elem() must be pointer to map.
  (applies to non-root as well)

unpriv: indirectly pass pointer on stack to helper function
  checks that pointer stored into stack cannot be used as part of key
  passed into bpf_map_lookup_elem()

unpriv: mangle pointer on stack 1
unpriv: mangle pointer on stack 2
  checks that writing into stack slot that already contains a pointer
  is disallowed

unpriv: read pointer from stack in small chunks
  checks that < 8 byte read from stack slot that contains a pointer is
  disallowed

unpriv: write pointer into ctx
  checks that storing pointers into skb->fields is disallowed

unpriv: write pointer into map elem value
  checks that storing pointers into element values is disallowed
  For example:
  int bpf_prog(struct __sk_buff *skb)
  {
    u32 key = 0;
    u64 *value = bpf_map_lookup_elem(&map, &key);
    if (value)
       *value = (u64) skb;
  }
  will be rejected.

unpriv: partial copy of pointer
  checks that doing 32-bit register mov from register containing
  a pointer is disallowed

unpriv: pass pointer to tail_call
  checks that passing pointer as an index into bpf_tail_call
  is disallowed

unpriv: cmp map pointer with zero
  checks that comparing map pointer with constant is disallowed

unpriv: write into frame pointer
  checks that frame pointer is read-only (applies to root too)

unpriv: cmp of frame pointer
  checks that R10 cannot be using in comparison

unpriv: cmp of stack pointer
  checks that Rx = R10 - imm is ok, but comparing Rx is not

unpriv: obfuscate stack pointer
  checks that Rx = R10 - imm is ok, but Rx -= imm is not
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

since eBPF programs and maps use kernel memory consider it 'locked' memory
from user accounting point of view and charge it against RLIMIT_MEMLOCK limit.
This limit is typically set to 64Kbytes by distros, so almost all
bpf+tracing programs would need to increase it, since they use maps,
but kernel charges maximum map size upfront.
For example the hash map of 1024 elements will be charged as 64Kbyte.
It's inconvenient for current users and changes current behavior for root,
but probably worth doing to be consistent root vs non-root.

Similar accounting logic is done by mmap of perf_event.
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>