openwrt/staging/blogic.git
8 years agonetfilter: nf_tables: allow to filter out rules by table and chain
Pablo Neira Ayuso [Tue, 19 Jul 2016 10:20:45 +0000 (12:20 +0200)]
netfilter: nf_tables: allow to filter out rules by table and chain

If the table and/or chain attributes are set in a rule dump request,
we filter out the rules based on this selection.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: nft_log: fix snaplen does not truncate packets
Liping Zhang [Mon, 18 Jul 2016 12:44:17 +0000 (20:44 +0800)]
netfilter: nft_log: fix snaplen does not truncate packets

There's a similar problem in xt_NFLOG, and was fixed by commit 7643507fe8b5
("netfilter: xt_NFLOG: nflog-range does not truncate packets"). Only set
copy_len here does not work, so we should enable NF_LOG_F_COPY_LEN also.

Signed-off-by: Liping Zhang <liping.zhang@spreadtrum.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: nft_log: check the validity of log level
Liping Zhang [Mon, 18 Jul 2016 12:44:16 +0000 (20:44 +0800)]
netfilter: nft_log: check the validity of log level

User can specify the log level larger than 7(debug level) via
nfnetlink, this is invalid. So in this case, we should report
EINVAL to the userspace.

Signed-off-by: Liping Zhang <liping.zhang@spreadtrum.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: nft_log: fix possible memory leak if log expr init fail
Liping Zhang [Mon, 18 Jul 2016 12:44:15 +0000 (20:44 +0800)]
netfilter: nft_log: fix possible memory leak if log expr init fail

Suppose that we specify the NFTA_LOG_PREFIX, then NFTA_LOG_LEVEL
and NFTA_LOG_GROUP are specified together or nf_logger_find_get
call returns fail, i.e. expr init fail, memory leak will happen.

Signed-off-by: Liping Zhang <liping.zhang@spreadtrum.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: Add helper array register/unregister functions
Gao Feng [Mon, 18 Jul 2016 03:39:23 +0000 (11:39 +0800)]
netfilter: Add helper array register/unregister functions

Add nf_ct_helper_init(), nf_conntrack_helpers_register() and
nf_conntrack_helpers_unregister() functions to avoid repetitive
opencoded initialization in helpers.

This patch keeps an id parameter for nf_ct_helper_init() not to break
helper matching by name that has been inconsistently exposed to
userspace through ports, eg. ftp-2121, and through an incremental id,
eg. tftp-1.

Signed-off-by: Gao Feng <fgao@ikuai8.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: nft_ct: fix unpaired nf_connlabels_get/put call
Liping Zhang [Sat, 16 Jul 2016 06:27:21 +0000 (14:27 +0800)]
netfilter: nft_ct: fix unpaired nf_connlabels_get/put call

We only get nf_connlabels if the user add ct label set expr successfully,
but we will also put nf_connlabels if the user delete ct lable get expr.
This is mismathced, and will cause ct label expr cannot work properly.

Also, if we init something fail, we should put nf_connlabels back.
Otherwise, we may waste to alloc the memory that will never be used.

Signed-off-by: Liping Zhang <liping.zhang@spreadtrum.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: x_tables: speed up jump target validation
Florian Westphal [Thu, 14 Jul 2016 15:51:26 +0000 (17:51 +0200)]
netfilter: x_tables: speed up jump target validation

The dummy ruleset I used to test the original validation change was broken,
most rules were unreachable and were not tested by mark_source_chains().

In some cases rulesets that used to load in a few seconds now require
several minutes.

sample ruleset that shows the behaviour:

echo "*filter"
for i in $(seq 0 100000);do
        printf ":chain_%06x - [0:0]\n" $i
done
for i in $(seq 0 100000);do
   printf -- "-A INPUT -j chain_%06x\n" $i
   printf -- "-A INPUT -j chain_%06x\n" $i
   printf -- "-A INPUT -j chain_%06x\n" $i
done
echo COMMIT

[ pipe result into iptables-restore ]

This ruleset will be about 74mbyte in size, with ~500k searches
though all 500k[1] rule entries. iptables-restore will take forever
(gave up after 10 minutes)

Instead of always searching the entire blob for a match, fill an
array with the start offsets of every single ipt_entry struct,
then do a binary search to check if the jump target is present or not.

After this change ruleset restore times get again close to what one
gets when reverting 36472341017529e (~3 seconds on my workstation).

[1] every user-defined rule gets an implicit RETURN, so we get
300k jumps + 100k userchains + 100k returns -> 500k rule entries

Fixes: 36472341017529e ("netfilter: x_tables: validate targets of jumps")
Reported-by: Jeff Wu <wujiafu@gmail.com>
Tested-by: Jeff Wu <wujiafu@gmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: conntrack: protect early_drop by rcu read lock
Liping Zhang [Tue, 12 Jul 2016 11:45:00 +0000 (19:45 +0800)]
netfilter: conntrack: protect early_drop by rcu read lock

User can add ct entry via nfnetlink(IPCTNL_MSG_CT_NEW), and if the total
number reach the nf_conntrack_max, we will try to drop some ct entries.

But in this case(the main function call path is ctnetlink_create_conntrack
-> nf_conntrack_alloc -> early_drop), rcu_read_lock is not held, so race
with hash resize will happen.

Fixes: 242922a02717 ("netfilter: conntrack: simplify early_drop")
Cc: Florian Westphal <fw@strlen.de>
Signed-off-by: Liping Zhang <liping.zhang@spreadtrum.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: nf_conntrack_h323: fix off-by-one in DecodeQ931
Toby DiPasquale [Mon, 11 Jul 2016 10:32:45 +0000 (11:32 +0100)]
netfilter: nf_conntrack_h323: fix off-by-one in DecodeQ931

This patch corrects an off-by-one error in the DecodeQ931 function in
the nf_conntrack_h323 module. This error could result in reading off
the end of a Q.931 frame.

Signed-off-by: Toby DiPasquale <toby@cbcg.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agoMerge tag 'ipvs-for-v4.8' of https://git.kernel.org/pub/scm/linux/kernel/git/horms...
Pablo Neira Ayuso [Mon, 11 Jul 2016 10:16:34 +0000 (12:16 +0200)]
Merge tag 'ipvs-for-v4.8' of https://git./linux/kernel/git/horms/ipvs-next

Simon Horman says:

====================
IPVS Updates for v4.8

please consider these enhancements to the IPVS. This alters the behaviour
of the "least connection" schedulers such that pre-established connections
are included in the active connection count. This avoids overloading
servers when a large number of new connections arrive in a short space of
time - e.g. when clients reconnect after a node or network failure.
====================

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: nf_tables: get rid of possible_net_t from set and basechain
Pablo Neira Ayuso [Fri, 8 Jul 2016 12:41:49 +0000 (14:41 +0200)]
netfilter: nf_tables: get rid of possible_net_t from set and basechain

We can pass the netns pointer as parameter to the functions that need to
gain access to it. From basechains, I didn't find any client for this
field anymore so let's remove this too.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: constify arg to is_dying/confirmed
Florian Westphal [Fri, 8 Jul 2016 11:08:50 +0000 (13:08 +0200)]
netfilter: constify arg to is_dying/confirmed

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: nft_ct: make byte/packet expr more friendly
Liping Zhang [Tue, 5 Jul 2016 15:23:00 +0000 (23:23 +0800)]
netfilter: nft_ct: make byte/packet expr more friendly

If we want to use ct packets expr, and add a rule like follows:
  # nft add rule filter input ct packets gt 1 counter

We will find that no packets will hit it, because
nf_conntrack_acct is disabled by default. So It will
not work until we enable it manually via
"echo 1 > /proc/sys/net/netfilter/nf_conntrack_acct".

This is not friendly, so like xt_connbytes do, if the user
want to use ct byte/packet expr, enable nf_conntrack_acct
automatically.

Signed-off-by: Liping Zhang <liping.zhang@spreadtrum.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: physdev: physdev-is-out should not work with OUTPUT chain
Hangbin Liu [Tue, 5 Jul 2016 12:55:36 +0000 (20:55 +0800)]
netfilter: physdev: physdev-is-out should not work with OUTPUT chain

physdev_mt() will check skb->nf_bridge first, which was alloced in
br_nf_pre_routing. So if we want to use --physdev-out and physdev-is-out,
we need to match it in FORWARD or POSTROUTING chain. physdev_mt_check()
only checked physdev-out and missed physdev-is-out. Fix it and update the
debug message to make it clearer.

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Marcelo R Leitner <marcelo.leitner@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: nat: convert nat bysrc hash to rhashtable
Florian Westphal [Tue, 5 Jul 2016 10:07:24 +0000 (12:07 +0200)]
netfilter: nat: convert nat bysrc hash to rhashtable

It did use a fixed-size bucket list plus single lock to protect add/del.

Unlike the main conntrack table we only need to add and remove keys.
Convert it to rhashtable to get table autosizing and per-bucket locking.

The maximum number of entries is -- as before -- tied to the number of
conntracks so we do not need another upperlimit.

The change does not handle rhashtable_remove_fast error, only possible
"error" is -ENOENT, and that is something that can happen legitimetely,
e.g. because nat module was inserted at a later time and no src manip
took place yet.

Tested with http-client-benchmark + httpterm with DNAT and SNAT rules
in place.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: move nat hlist_head to nf_conn
Florian Westphal [Tue, 5 Jul 2016 10:07:23 +0000 (12:07 +0200)]
netfilter: move nat hlist_head to nf_conn

The nat extension structure is 32bytes in size on x86_64:

struct nf_conn_nat {
        struct hlist_node          bysource;             /*     0    16 */
        struct nf_conn *           ct;                   /*    16     8 */
        union nf_conntrack_nat_help help;                /*    24     4 */
        int                        masq_index;           /*    28     4 */
        /* size: 32, cachelines: 1, members: 4 */
        /* last cacheline: 32 bytes */
};

The hlist is needed to quickly check for possible tuple collisions
when installing a new nat binding. Storing this in the extension
area has two drawbacks:

1. We need ct backpointer to get the conntrack struct from the extension.
2. When reallocation of extension area occurs we need to fixup the bysource
   hash head via hlist_replace_rcu.

We can avoid both by placing the hlist_head in nf_conn and place nf_conn in
the bysource hash rather than the extenstion.

We can also remove the ->move support; no other extension needs it.

Moving the entire nat extension into nf_conn would be possible as well but
then we have to add yet another callback for deletion from the bysource
hash table rather than just using nat extension ->destroy hook for this.

nf_conn size doesn't increase due to aligment, followup patch replaces
hlist_node with single pointer.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: conntrack: simplify early_drop
Florian Westphal [Sun, 3 Jul 2016 18:44:01 +0000 (20:44 +0200)]
netfilter: conntrack: simplify early_drop

We don't need to acquire the bucket lock during early drop, we can
use lockless traveral just like ____nf_conntrack_find.

The timer deletion serves as synchronization point, if another cpu
attempts to evict same entry, only one will succeed with timer deletion.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: nf_ct_helper: unlink helper again when hash resize happen
Liping Zhang [Sun, 3 Jul 2016 05:18:45 +0000 (13:18 +0800)]
netfilter: nf_ct_helper: unlink helper again when hash resize happen

From: Liping Zhang <liping.zhang@spreadtrum.com>

Similar to ctnl_untimeout, when hash resize happened, we should try
to do unhelp from the 0# bucket again.

Signed-off-by: Liping Zhang <liping.zhang@spreadtrum.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: cttimeout: unlink timeout obj again when hash resize happen
Liping Zhang [Sun, 3 Jul 2016 05:18:44 +0000 (13:18 +0800)]
netfilter: cttimeout: unlink timeout obj again when hash resize happen

Imagine such situation, nf_conntrack_htable_size now is 4096, we are doing
ctnl_untimeout, and iterate on 3000# bucket.

Meanwhile, another user try to reduce hash size to 2048, then all nf_conn
are removed to the new hashtable. When this hash resize operation finished,
we still try to itreate ct begin from 3000# bucket, find nothing to do and
just return.

We may miss unlinking some timeout objects. And later we will end up with
invalid references to timeout object that are already gone.

So when we find that hash resize happened, try to unlink timeout objects
from the 0# bucket again.

Signed-off-by: Liping Zhang <liping.zhang@spreadtrum.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agonetfilter: conntrack: fix race between nf_conntrack proc read and hash resize
Liping Zhang [Sun, 3 Jul 2016 05:18:43 +0000 (13:18 +0800)]
netfilter: conntrack: fix race between nf_conntrack proc read and hash resize

When we do "cat /proc/net/nf_conntrack", and meanwhile resize the conntrack
hash table via /sys/module/nf_conntrack/parameters/hashsize, race will
happen, because reader can observe a newly allocated hash but the old size
(or vice versa). So oops will happen like follows:

  BUG: unable to handle kernel NULL pointer dereference at 0000000000000017
  IP: [<ffffffffa0418e21>] seq_print_acct+0x11/0x50 [nf_conntrack]
  Call Trace:
  [<ffffffffa0412f4e>] ? ct_seq_show+0x14e/0x340 [nf_conntrack]
  [<ffffffff81261a1c>] seq_read+0x2cc/0x390
  [<ffffffff812a8d62>] proc_reg_read+0x42/0x70
  [<ffffffff8123bee7>] __vfs_read+0x37/0x130
  [<ffffffff81347980>] ? security_file_permission+0xa0/0xc0
  [<ffffffff8123cf75>] vfs_read+0x95/0x140
  [<ffffffff8123e475>] SyS_read+0x55/0xc0
  [<ffffffff817c2572>] entry_SYSCALL_64_fastpath+0x1a/0xa4

It is very easy to reproduce this kernel crash.
1. open one shell and input the following cmds:
  while : ; do
    echo $RANDOM > /sys/module/nf_conntrack/parameters/hashsize
  done
2. open more shells and input the following cmds:
  while : ; do
    cat /proc/net/nf_conntrack
  done
3. just wait a monent, oops will happen soon.

The solution in this patch is based on Florian's Commit 5e3c61f98175
("netfilter: conntrack: fix lookup race during hash resize"). And
add a wrapper function nf_conntrack_get_ht to get hash and hsize
suggested by Florian Westphal.

Signed-off-by: Liping Zhang <liping.zhang@spreadtrum.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agoipvs: count pre-established TCP states as active
Michal Kubecek [Fri, 3 Jun 2016 15:56:50 +0000 (17:56 +0200)]
ipvs: count pre-established TCP states as active

Some users observed that "least connection" distribution algorithm doesn't
handle well bursts of TCP connections from reconnecting clients after
a node or network failure.

This is because the algorithm counts active connection as worth 256
inactive ones where for TCP, "active" only means TCP connections in
ESTABLISHED state. In case of a connection burst, new connections are
handled before previous ones have finished the three way handshaking so
that all are still counted as "inactive", i.e. cheap ones. The become
"active" quickly but at that time, all of them are already assigned to one
real server (or few), resulting in highly unbalanced distribution.

Address this by counting the "pre-established" states as "active".

Signed-off-by: Michal Kubecek <mkubecek@suse.cz>
Acked-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
8 years agoMerge tag 'mac80211-next-for-davem-2016-07-06' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Thu, 7 Jul 2016 05:32:15 +0000 (22:32 -0700)]
Merge tag 'mac80211-next-for-davem-2016-07-06' of git://git./linux/kernel/git/jberg/mac80211-next

Johannes Berg says:

====================
One more set of new features:
 * beacon report (for radio measurement) support in cfg80211/mac80211
 * hwsim: allow wmediumd in namespaces
 * mac80211: extend 160MHz workaround to CSA IEs
 * mesh: properly encrypt group-addressed privacy action frames
 * mesh: allow setting peer AID
 * first steps for MU-MIMO monitor mode
 * along with various other cleanups and improvements
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: mediatek: remove .owner field for driver
Wei Yongjun [Wed, 6 Jul 2016 13:30:13 +0000 (13:30 +0000)]
net: mediatek: remove .owner field for driver

Remove .owner field since calls to module_platform_driver() will
set it automatically.

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Wed, 6 Jul 2016 17:35:22 +0000 (10:35 -0700)]
Merge git://git./linux/kernel/git/davem/net

Conflicts:
drivers/net/ethernet/mellanox/mlx5/core/en.h
drivers/net/ethernet/mellanox/mlx5/core/en_main.c
drivers/net/usb/r8152.c

All three conflicts were overlapping changes.

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Wed, 6 Jul 2016 16:42:43 +0000 (09:42 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) All users of AF_PACKET's fanout feature want a symmetric packet
    header hash for load balancing purposes, so give it to them.

 2) Fix vlan state synchronization in e1000e, from Jarod Wilson.

 3) Use correct socket pointer in ip_skb_dst_mtu(), from Shmulik
    Ladkani.

 4) mlx5 bug fixes from Mohamad Haj Yahia, Daniel Jurgens, Matthew
    Finlay, Rana Shahout, and Shaker Daibes.  Mostly to do with
    operation timeouts and PCI error handling.

 5) Fix checksum handling in mirred packet action, from WANG Cong.

 6) Set skb->dev correctly when transmitting in !protect_frames case of
    macsec driver, from Daniel Borkmann.

 7) Fix MTU calculation in geneve driver, from Haishuang Yan.

 8) Missing netif_napi_del() in unregister path of qeth driver, from
    Ursula Braun.

 9) Handle malformed route netlink messages in decnet properly, from
    Vergard Nossum.

10) Memory leak of percpu data in ipv6 routing code, from Martin KaFai
    Lau.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (41 commits)
  ipv6: Fix mem leak in rt6i_pcpu
  net: fix decnet rtnexthop parsing
  cxgb4: update latest firmware version supported
  net/mlx5: Avoid setting unused var when modifying vport node GUID
  bonding: fix enslavement slave link notifications
  r8152: fix runtime function for RTL8152
  qeth: delete napi struct when removing a qeth device
  Revert "fsl/fman: fix error handling"
  fsl/fman: fix error handling
  cdc_ncm: workaround for EM7455 "silent" data interface
  RDS: fix rds_tcp_init() error path
  geneve: fix max_mtu setting
  net: phy: dp83867: Fix initialization of PHYCR register
  enc28j60: Fix race condition in enc28j60 driver
  net: stmmac: Fix null-function call in ISR on stmmac1000
  tipc: fix nl compat regression for link statistics
  net: bcmsysport: Device stats are unsigned long
  macsec: set actual real device for xmit when !protect_frames
  net_sched: fix mirrored packets checksum
  packet: Use symmetric hash for PACKET_FANOUT_HASH.
  ...

8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next
David S. Miller [Wed, 6 Jul 2016 16:15:15 +0000 (09:15 -0700)]
Merge git://git./linux/kernel/git/pablo/nf-next

Pablo Neira Ayuso says:

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

The following patchset contains Netfilter updates for net-next,
they are:

1) Don't use userspace datatypes in bridge netfilter code, from
   Tobin Harding.

2) Iterate only once over the expectation table when removing the
   helper module, instead of once per-netns, from Florian Westphal.

3) Extra sanitization in xt_hook_ops_alloc() to return error in case
   we ever pass zero hooks, xt_hook_ops_alloc():

4) Handle NFPROTO_INET from the logging core infrastructure, from
   Liping Zhang.

5) Autoload loggers when TRACE target is used from rules, this doesn't
   change the behaviour in case the user already selected nfnetlink_log
   as preferred way to print tracing logs, also from Liping Zhang.

6) Conntrack slabs with SLAB_HWCACHE_ALIGN to allow rearranging fields
   by cache lines, increases the size of entries in 11% per entry.
   From Florian Westphal.

7) Skip zone comparison if CONFIG_NF_CONNTRACK_ZONES=n, from Florian.

8) Remove useless defensive check in nf_logger_find_get() from Shivani
   Bhardwaj.

9) Remove zone extension as place it in the conntrack object, this is
   always include in the hashing and we expect more intensive use of
   zones since containers are in place. Also from Florian Westphal.

10) Owner match now works from any namespace, from Eric Bierdeman.

11) Make sure we only reply with TCP reset to TCP traffic from
    nf_reject_ipv4, patch from Liping Zhang.

12) Introduce --nflog-size to indicate amount of network packet bytes
    that are copied to userspace via log message, from Vishwanath Pai.
    This obsoletes --nflog-range that has never worked, it was designed
    to achieve this but it has never worked.

13) Introduce generic macros for nf_tables object generation masks.

14) Use generation mask in table, chain and set objects in nf_tables.
    This allows fixes interferences with ongoing preparation phase of
    the commit protocol and object listings going on at the same time.
    This update is introduced in three patches, one per object.

15) Check if the object is active in the next generation for element
    deactivation in the rbtree implementation, given that deactivation
    happens from the commit phase path we have to observe the future
    status of the object.

16) Support for deletion of just added elements in the hash set type.

17) Allow to resize hashtable from /proc entry, not only from the
    obscure /sys entry that maps to the module parameter, from Florian
    Westphal.

18) Get rid of NFT_BASECHAIN_DISABLED, this code is not exercised
    anymore since we tear down the ruleset whenever the netdevice
    goes away.

19) Support for matching inverted set lookups, from Arturo Borrero.

20) Simplify the iptables_mangle_hook() by removing a superfluous
    extra branch.

21) Introduce ether_addr_equal_masked() and use it from the netfilter
    codebase, from Joe Perches.

22) Remove references to "Use netfilter MARK value as routing key"
    from the Netfilter Kconfig description given that this toggle
    doesn't exists already for 10 years, from Moritz Sichert.

23) Introduce generic NF_INVF() and use it from the xtables codebase,
    from Joe Perches.

24) Setting logger to NONE via /proc was not working unless explicit
    nul-termination was included in the string. This fixes seems to
    leave the former behaviour there, so we don't break backward.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'sound-4.7-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
Linus Torvalds [Wed, 6 Jul 2016 16:12:43 +0000 (09:12 -0700)]
Merge tag 'sound-4.7-rc7' of git://git./linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "Here are a collection of small fixes: at this time, we've got a
  slightly high amount, but all small and trivial fixes, and nothing
  scary can be seen there"

* tag 'sound-4.7-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound: (21 commits)
  ALSA: hda/realtek: Add Lenovo L460 to docking unit fixup
  ALSA: timer: Fix negative queue usage by racy accesses
  ASoC: rt5645: fix reg-2f default value.
  ASoC: fsl_ssi: Fix number of words per frame for I2S-slave mode
  ALSA: au88x0: Fix calculation in vortex_wtdma_bufshift()
  ALSA: hda - Add PCI ID for Kabylake-H
  ALSA: echoaudio: Fix memory allocation
  ASoC: Intel: atom: fix missing breaks that would cause the wrong operation to execute
  ALSA: hda - fix read before array start
  ASoC: cx20442: set tty->receiver_room in v253_open
  ASoC: ak4613: Enable cache usage to fix crashes on resume
  ASoC: wm8940: Enable cache usage to fix crashes on resume
  ASoC: Intel: Skylake: Initialize module list for Broxton
  ASoC: wm5102: Correct supported channels on trace compressed DAI
  ASoC: wm5110: Add missing route from OUT3R to SYSCLK
  ASoC: rt5670: fix HP Playback Volume control
  ASoC: hdmi-codec: select CONFIG_HDMI
  ASoC: davinci-mcasp: Fix dra7 DMA offset when using CFG port
  ASoC: hdac_hdmi: Fix potential NULL dereference
  ASoC: ak4613: Remove owner assignment from platform_driver
  ...

8 years agoMerge tag 'chrome-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/olof/chrome...
Linus Torvalds [Wed, 6 Jul 2016 16:07:23 +0000 (09:07 -0700)]
Merge tag 'chrome-fixes' of git://git./linux/kernel/git/olof/chrome-platform

Pull chrome platform fix from Olof Johansson:
 "A single fix this time, closing a window where ioctl args are fetched
  twice"

* tag 'chrome-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/olof/chrome-platform:
  platform/chrome: cros_ec_dev - double fetch bug in ioctl

8 years agocfg80211: Add mesh peer AID setting API
Masashi Honma [Fri, 1 Jul 2016 01:19:34 +0000 (10:19 +0900)]
cfg80211: Add mesh peer AID setting API

Previously, mesh power management functionality works only with kernel
MPM. Because user space MPM did not report mesh peer AID to kernel,
the kernel could not identify the bit in TIM element. So this patch
adds mesh peer AID setting API.

Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
8 years agomac80211: parse wide bandwidth channel switch IE with workaround
Johannes Berg [Tue, 5 Jul 2016 12:23:14 +0000 (15:23 +0300)]
mac80211: parse wide bandwidth channel switch IE with workaround

Continuing the workaround implemented in commit 23665aaf9170
("mac80211: Interoperability workaround for 80+80 and 160 MHz channels")
use the same code to parse the Wide Bandwidth Channel Switch element
by converting to VHT Operation element since the spec also just refers
to that for parsing semantics, particularly with the workaround.

While at it, remove some dead code - the IEEE80211_STA_DISABLE_40MHZ
flag can never be set at this point since it's checked earlier and the
wide_bw_chansw_ie pointer is set to NULL if it's set.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
8 years agomac80211: report failure to start (partial) scan as scan abort
Johannes Berg [Tue, 5 Jul 2016 12:23:13 +0000 (15:23 +0300)]
mac80211: report failure to start (partial) scan as scan abort

Rather than reporting the scan as having completed, report it as
being aborted.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
8 years agomac80211: Add support for beacon report radio measurement
Avraham Stern [Tue, 5 Jul 2016 12:23:12 +0000 (15:23 +0300)]
mac80211: Add support for beacon report radio measurement

Add the following to support beacon report radio measurement
with the measurement mode field set to passive or active:
1. Propagate the required scan duration to the device
2. Report the scan start time (in terms of TSF)
3. Report each BSS's detection time (also in terms of TSF)

TSF times refer to the BSS that the interface that requested the
scan is connected to.

Signed-off-by: Assaf Krauss <assaf.krauss@intel.com>
Signed-off-by: Avraham Stern <avraham.stern@intel.com>
[changed ath9k/10k, at76c59x-usb, iwlegacy, wl1251 and wlcore to match
the new API]
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
8 years agonl80211: support beacon report scanning
Avraham Stern [Tue, 5 Jul 2016 14:10:13 +0000 (17:10 +0300)]
nl80211: support beacon report scanning

Beacon report radio measurement requires reporting observed BSSs
on the channels specified in the beacon request. If the measurement
mode is set to passive or active, it requires actually performing a
scan (passive or active, accordingly), and reporting the time that
the scan was started and the time each beacon/probe was received
(both in terms of TSF of the BSS of the requesting AP). If the
request mode is table, this information is optional.
In addition, the radio measurement request specifies the channel
dwell time for the measurement.

In order to use scan for beacon report when the mode is active or
passive, add a parameter to scan request that specifies the
channel dwell time, and add scan start time and beacon received time
to scan results information.

Supporting beacon report is required for Multi Band Operation (MBO).

Signed-off-by: Assaf Krauss <assaf.krauss@intel.com>
Signed-off-by: David Spinadel <david.spinadel@intel.com>
Signed-off-by: Avraham Stern <avraham.stern@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
8 years agomac80211_hwsim: use signed net namespace ID
Johannes Berg [Wed, 6 Jul 2016 12:47:49 +0000 (14:47 +0200)]
mac80211_hwsim: use signed net namespace ID

The API expects a pointer to a signed int so we should not use an
unsigned int for it.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
8 years agomac80211_hwsim: Add radar bandwidths to the P2P Device combination
Ilan Peer [Tue, 5 Jul 2016 12:23:09 +0000 (15:23 +0300)]
mac80211_hwsim: Add radar bandwidths to the P2P Device combination

Add radar_detect_widths to the interface combination that allows
concurrent P2P Device dedicated interface and AP interfaces, to enable
testing of radar detection when P2P Device interface is used.

Clear the radar_detect_widths in case of multi channel contexts
as this is not currently supported.

As radar_detect_widths are now supported in all combinations,
remove the hwsim_if_dfs_limits definition since it is no longer
needed.

Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
8 years agonl80211: Add API to support VHT MU-MIMO air sniffer
Aviya Erenfeld [Tue, 5 Jul 2016 12:23:08 +0000 (15:23 +0300)]
nl80211: Add API to support VHT MU-MIMO air sniffer

add API to support VHT MU-MIMO air sniffer.
in MU-MIMO there are parallel frames on the air while the HW
has only one RX.
add the capability to sniff one of the MU-MIMO parallel frames by
giving the sniffer additional information so it'll know which
of the parallel frames it shall follow.

Add attribute - NL80211_ATTR_MU_MIMO_GROUP_DATA - for getting
a MU-MIMO groupID in order to monitor packets from that group
using VHT MU-MIMO.
And add attribute -NL80211_ATTR_MU_MIMO_FOLLOW_ADDR - for passing
MAC address to monitor mode.
that option will be used by VHT MU-MIMO air sniffer to follow a
station according to it's MAC address using VHT MU-MIMO.

Signed-off-by: Aviya Erenfeld <aviya.erenfeld@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
8 years agomac80211: agg-rx: refuse ADDBA Request with timeout update
Johannes Berg [Wed, 6 Jul 2016 12:44:14 +0000 (14:44 +0200)]
mac80211: agg-rx: refuse ADDBA Request with timeout update

The current implementation of handling ADDBA Request while a session
is already active with the peer is wrong - in case the peer is using
the existing session's dialog token this should be treated as update
to the session, which can update the timeout value.

We don't really have a good way of supporting that, so reject, but
implement the required behaviour in the spec of "Even if the updated
ADDBA Request frame is not accepted, the original Block ACK setup
remains active." (802.11-2012 10.5.4)

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
8 years agoipv6: Fix mem leak in rt6i_pcpu
Martin KaFai Lau [Tue, 5 Jul 2016 19:10:23 +0000 (12:10 -0700)]
ipv6: Fix mem leak in rt6i_pcpu

It was first reported and reproduced by Petr (thanks!) in
https://bugzilla.kernel.org/show_bug.cgi?id=119581

free_percpu(rt->rt6i_pcpu) used to always happen in ip6_dst_destroy().

However, after fixing a deadlock bug in
commit 9c7370a166b4 ("ipv6: Fix a potential deadlock when creating pcpu rt"),
free_percpu() is not called before setting non_pcpu_rt->rt6i_pcpu to NULL.

It is worth to note that rt6i_pcpu is protected by table->tb6_lock.

kmemleak somehow did not report it.  We nailed it down by
observing the pcpu entries in /proc/vmallocinfo (first suggested
by Hannes, thanks!).

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Fixes: 9c7370a166b4 ("ipv6: Fix a potential deadlock when creating pcpu rt")
Reported-by: Petr Novopashenniy <pety@rusnet.ru>
Tested-by: Petr Novopashenniy <pety@rusnet.ru>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Petr Novopashenniy <pety@rusnet.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: fix decnet rtnexthop parsing
Vegard Nossum [Tue, 5 Jul 2016 19:12:53 +0000 (21:12 +0200)]
net: fix decnet rtnexthop parsing

dn_fib_count_nhs() could enter an infinite loop if nhp->rtnh_len == 0
(i.e. if userspace passes a malformed netlink message).

Let's use the helpers from net/nexthop.h which take care of all this
stuff. We can do exactly the same as e.g. fib_count_nexthops() and
fib_get_nhs() from net/ipv4/fib_semantics.c.

This fixes the softlockup for me.

Cc: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Vegard Nossum <vegard.nossum@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoplatform/chrome: cros_ec_dev - double fetch bug in ioctl
Dan Carpenter [Tue, 21 Jun 2016 13:58:46 +0000 (16:58 +0300)]
platform/chrome: cros_ec_dev - double fetch bug in ioctl

We verify "u_cmd.outsize" and "u_cmd.insize" but we need to make sure
that those values have not changed between the two copy_from_user()
calls.  Otherwise it could lead to a buffer overflow.

Additionally, cros_ec_cmd_xfer() can set s_cmd->insize to a lower value.
We should use the new smaller value so we don't copy too much data to
the user.

Reported-by: Pengfei Wang <wpengfeinudt@gmail.com>
Fixes: a841178445bb ('mfd: cros_ec: Use a zero-length array for command data')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Tested-by: Gwendal Grignou <gwendal@chromium.org>
Cc: <stable@vger.kernel.org> # v4.2+
Signed-off-by: Olof Johansson <olof@lixom.net>
8 years agocxgb4: update latest firmware version supported
Ganesh Goudar [Tue, 5 Jul 2016 12:37:24 +0000 (18:07 +0530)]
cxgb4: update latest firmware version supported

Change t4fw_version.h to update latest firmware version number

Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5: Avoid setting unused var when modifying vport node GUID
Or Gerlitz [Tue, 5 Jul 2016 09:17:12 +0000 (12:17 +0300)]
net/mlx5: Avoid setting unused var when modifying vport node GUID

GCC complains on unused-but-set-variable, clean this up.

Fixes: 23898c763f4a ('net/mlx5: E-Switch, Modify node guid on vf set MAC')
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobonding: fix enslavement slave link notifications
Aviv Heller [Tue, 5 Jul 2016 09:09:47 +0000 (12:09 +0300)]
bonding: fix enslavement slave link notifications

Currently, link notifications are not sent by
bond_set_slave_link_state() upon enslavement if
the slave is enslaved when up.

This happens because slave->link default init value
is 0, which is the same as BOND_LINK_UP, resulting
in bond_set_slave_link_state() ignoring this transition.

This patch sets the default value of slave->link to
BOND_LINK_NOCHANGE, assuring it will count as a state
transition and thus trigger notification logic.

Signed-off-by: Aviv Heller <avivh@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: vrf: Add support for PREROUTING rules on vrf device
David Ahern [Tue, 5 Jul 2016 01:47:41 +0000 (18:47 -0700)]
net: vrf: Add support for PREROUTING rules on vrf device

Add support for PREROUTING rules with skb->dev set to the vrf device.
INPUT rules are already allowed. Provides symmetry with the output path
which allows POSTROUTING rules.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoconnector: make cn_proc explicitly non-modular
Paul Gortmaker [Mon, 4 Jul 2016 21:50:58 +0000 (17:50 -0400)]
connector: make cn_proc explicitly non-modular

The Kconfig controlling build of this code is currently:

drivers/connector/Kconfig:config PROC_EVENTS
drivers/connector/Kconfig:      bool "Report process events to userspace"

...meaning that it currently is not being built as a module by anyone.
Lets remove the two modular references, so that when reading the driver
there is no doubt it is builtin-only.

Since module_init translates to device_initcall in the non-modular
case, the init ordering remains unchanged with this commit.

Cc: Evgeniy Polyakov <zbr@ioremap.net>
Cc: netdev@vger.kernel.org
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agor8152: fix runtime function for RTL8152
hayeswang [Tue, 5 Jul 2016 08:11:46 +0000 (16:11 +0800)]
r8152: fix runtime function for RTL8152

The RTL8152 doesn't have U1U2 and U2P3 features, so use different
runtime functions for RTL812 and RTL8153 by adding autosuspend_en()
to rtl_ops.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: fix return value check in hns_dsaf_get_cfg()
Wei Yongjun [Tue, 5 Jul 2016 07:56:52 +0000 (07:56 +0000)]
net: hns: fix return value check in hns_dsaf_get_cfg()

In case of error, function devm_ioremap_resource() returns ERR_PTR()
and never returns NULL. The NULL test in the return value check should
be replaced with IS_ERR().

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'mlxsw-ipv4-unicast-routing'
David S. Miller [Tue, 5 Jul 2016 16:06:31 +0000 (09:06 -0700)]
Merge branch 'mlxsw-ipv4-unicast-routing'

Jiri Pirko says:

====================
mlxsw: Implement IPV4 unicast routing

This patchset enables IPv4 unicast routing in the Mellanox Spectrum ASIC
switch driver. This builds upon the work that was done by a couple of
previous patchsets.

Patches 1,2,6 add a couple of dependencies outside the driver. Namely, the
ability to propagate ndo_neigh_construct()/destroy() through stacked devices and
a notification whenever DELAY_PROBE_TIME changes. When propagated down, the
ndos allow drivers to add and remove neighbour entries from their private
neighbour table. The DELAY_PROBE_TIME notification gives drivers the ability to
correctly configure their polling interval for neighbour activity, so that
active neighbour won't be marked as STALE.

Patches 3-5,7-8 add the neighbour offloading infrastructure, where patch 7 uses
the DELAY_PROBE_TIME notification in order to correctly configure the device's
polling interval. Patch 8 finally programs neighbours to the device's table
based on NEIGH_UPDATE notifications, so that directly connected routes can
be used.

Patches 9-16 build upon the previous patches and extend the router with
remote routes (nexthop) support.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: Add the unresolved next-hops probes
Yotam Gigi [Tue, 5 Jul 2016 09:27:52 +0000 (11:27 +0200)]
mlxsw: Add the unresolved next-hops probes

Now, the driver sends arp probes for all unresolved neighbours that are
currently a nexthop for some route on the system. The job is set
periodically every 5 seconds.

Signed-off-by: Yotam Gigi <yotamg@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum_router: Add the nexthop neigh activity update
Yotam Gigi [Tue, 5 Jul 2016 09:27:51 +0000 (11:27 +0200)]
mlxsw: spectrum_router: Add the nexthop neigh activity update

For nexthop neighbours we need to make kernel to think there is a traffic
flowing to them preventing it from going to stale state. Otherwise
kernel would stale it and eventually the neigh would be removed from HW
and nexthop as well. That would reduce ECMP group in HW.

Signed-off-by: Yotam Gigi <yotamg@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum_router: Implement next-hop routing
Jiri Pirko [Tue, 5 Jul 2016 09:27:50 +0000 (11:27 +0200)]
mlxsw: spectrum_router: Implement next-hop routing

Implement next-hop routing offload including ECMP. To make it possible,
introduce next-hop group entity. This entity keeps track of resolved
neighbours and updates HW adjacency table accordingly. Note that HW
next-hops are stored in this adjacency table, in form of MAC.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: reg: Add Router Algorithmic LPM ECMP Update Register
Jiri Pirko [Tue, 5 Jul 2016 09:27:49 +0000 (11:27 +0200)]
mlxsw: reg: Add Router Algorithmic LPM ECMP Update Register

The RALEU register is used to mass update remote action adjacency index
and ecmp size.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: reg: Add Router Adjacency Table register
Yotam Gigi [Tue, 5 Jul 2016 09:27:48 +0000 (11:27 +0200)]
mlxsw: reg: Add Router Adjacency Table register

The RATR register is used to configure the Router Adjacency (next-hop)
Table.

Signed-off-by: Yotam Gigi <yotamg@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: Introduce simplistic KVD linear area manager
Jiri Pirko [Tue, 5 Jul 2016 09:27:47 +0000 (11:27 +0200)]
mlxsw: Introduce simplistic KVD linear area manager

This is a very simple manager for KVD linear area. Currently, the
allocator will either allocate a single entry from pre-defined sub-area,
or in case more than one entry is needed, it will allocate 32-entry chunk
in other pre-defined sub-area.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Define sizes of KVD areas
Jiri Pirko [Tue, 5 Jul 2016 09:27:46 +0000 (11:27 +0200)]
mlxsw: spectrum: Define sizes of KVD areas

Override the defaults and define the area sizes ourselves.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: Add KVD sizes configuration into profile
Jiri Pirko [Tue, 5 Jul 2016 09:27:45 +0000 (11:27 +0200)]
mlxsw: Add KVD sizes configuration into profile

Up until now we only used hash-based tables in the device, but we are
going to use the linear table for remote routes adjacency lists.

Add the configuration fields that control the size of the linear table.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum_router: Offload neighbours based on NUD state change
Yotam Gigi [Tue, 5 Jul 2016 09:27:44 +0000 (11:27 +0200)]
mlxsw: spectrum_router: Offload neighbours based on NUD state change

Listen to any NEIGH_UPDATE events sent and program the device
accordingly. If NUD state is VALID and neighbour isn't yet offloaded,
then program it into the device's table. Otherwise, just edit its
parameters.

If NUD state machine transitioned neighbour out of VALID state and it's
present in the device's table, then remove it.

Note that the device is programmed in delayed work, as the netevent
notification chain is atomic and prevents us from going to sleep.

Signed-off-by: Yotam Gigi <yotamg@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum_router: Periodically update the kernel's neigh table
Yotam Gigi [Tue, 5 Jul 2016 09:27:43 +0000 (11:27 +0200)]
mlxsw: spectrum_router: Periodically update the kernel's neigh table

As previously explained, the driver should periodically poll the device
for neighbours activity according to the configured DELAY_PROBE_TIME.
This will prevent active neighbours from staying in STALE state for long
periods of time.

During init configure the polling interval according to the
DELAY_PROBE_TIME used in the default table. In addition, register a
netevent notification block, so that the interval is updated whenever
DELAY_PROBE_TIME changes.

Using the computed interval schedule a delayed work, which will update
the kernel via neigh_event_send() on any active neighbour since the last
delayed work.

Signed-off-by: Yotam Gigi <yotamg@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoneigh: Send a notification when DELAY_PROBE_TIME changes
Ido Schimmel [Tue, 5 Jul 2016 09:27:42 +0000 (11:27 +0200)]
neigh: Send a notification when DELAY_PROBE_TIME changes

When the data plane is offloaded the traffic doesn't go through the
networking stack. Therefore, after first resolving a neighbour the NUD
state machine will transition it from REACHABLE to STALE until it's
finally deleted by the garbage collector.

To prevent such situations the offloading driver should notify the NUD
state machine on any neighbours that were recently used. The driver's
polling interval should be set so that the NUD state machine can
function as if the traffic wasn't offloaded.

Currently, there are no in-tree drivers that can report confirmation for
a neighbour, but only 'used' indication. Therefore, the polling interval
should be set according to DELAY_FIRST_PROBE_TIME, as a neighbour will
transition from REACHABLE state to DELAY (instead of STALE) if "a packet
was sent within the last DELAY_FIRST_PROBE_TIME seconds" (RFC 4861).

Send a netevent whenever the DELAY_FIRST_PROBE_TIME changes - either via
netlink or sysctl - so that offloading drivers can correctly set their
polling interval.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: reg: Add Router Algorithmic LPM Unicast Host Table Dump register
Yotam Gigi [Tue, 5 Jul 2016 09:27:41 +0000 (11:27 +0200)]
mlxsw: reg: Add Router Algorithmic LPM Unicast Host Table Dump register

The RAUHTD register allows dumping entries from the Router Unicast Host
Table.

Signed-off-by: Yotam Gigi <yotamg@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: reg: Add Router Algorithmic LPM Unicast Host Table register
Yotam Gigi [Tue, 5 Jul 2016 09:27:40 +0000 (11:27 +0200)]
mlxsw: reg: Add Router Algorithmic LPM Unicast Host Table register

The RAUHT register is used to configure and query the Unicast Host Table
in devices that implement the Algorithmic LPM. In other words, it is
used to configure neighbour entries in the device.

Signed-off-by: Yotam Gigi <yotamg@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum_router: Add private neigh table
Jiri Pirko [Tue, 5 Jul 2016 09:27:39 +0000 (11:27 +0200)]
mlxsw: spectrum_router: Add private neigh table

We need to hold some private data for every neigh entry. It would be
possible to do it using neigh_priv_len/ndo_neigh_construct/
ndo_neigh_destroy however only for the port device itself. That would not
work for stacked devices like bridge/team/bond. So introduce a private
neigh table. Hook onto ndos neigh_construct/destroy and add/remove
table entry according to that.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: introduce default neigh_construct/destroy ndo calls for L2 upper devices
Jiri Pirko [Tue, 5 Jul 2016 09:27:38 +0000 (11:27 +0200)]
net: introduce default neigh_construct/destroy ndo calls for L2 upper devices

L2 upper device needs to propagate neigh_construct/destroy calls down to
lower devices. Do this by defining default ndo functions and use them in
team, bond, bridge and vlan.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: add dev arg to ndo_neigh_construct/destroy
Jiri Pirko [Tue, 5 Jul 2016 09:27:37 +0000 (11:27 +0200)]
net: add dev arg to ndo_neigh_construct/destroy

As the following patch will allow upper devices to follow the call down
lower devices, we need to add dev here and not rely on n->dev.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonetfilter: nf_log: fix error on write NONE to logger choice sysctl
Pavel Tikhomirov [Fri, 1 Jul 2016 13:53:54 +0000 (16:53 +0300)]
netfilter: nf_log: fix error on write NONE to logger choice sysctl

It is hard to unbind nf-logger:

  echo NONE > /proc/sys/net/netfilter/nf_log/0
  bash: echo: write error: No such file or directory

  sysctl -w net.netfilter.nf_log.0=NONE
  sysctl: setting key "net.netfilter.nf_log.0": No such file or directory
  net.netfilter.nf_log.0 = NONE

You need explicitly send '\0', for instance like:

  echo -e "NONE\0" > /proc/sys/net/netfilter/nf_log/0

That seem to be strange, so fix it using proc_dostring.

Now it works fine:
   modprobe nfnetlink_log
   echo nfnetlink_log > /proc/sys/net/netfilter/nf_log/0
   cat /proc/sys/net/netfilter/nf_log/0
   nfnetlink_log
   echo NONE > /proc/sys/net/netfilter/nf_log/0
   cat /proc/sys/net/netfilter/nf_log/0
   NONE

v2: add missed error check for proc_dostring

Signed-off-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
8 years agoALSA: hda/realtek: Add Lenovo L460 to docking unit fixup
Torsten Hilbrich [Tue, 5 Jul 2016 08:40:22 +0000 (10:40 +0200)]
ALSA: hda/realtek: Add Lenovo L460 to docking unit fixup

This solves the issue that a headphone is not working on the docking
unit.

Signed-off-by: Torsten Hilbrich <torsten.hilbrich@secunet.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
8 years agoMerge branch 'r6040-next'
David S. Miller [Tue, 5 Jul 2016 07:10:30 +0000 (00:10 -0700)]
Merge branch 'r6040-next'

Florian Fainelli says:

====================
net: r6040: Misc updates

Here are some various updates for the r6040 driver, mostly to make it more
modern and catch up with the latest API improvements.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: r6040: Bump version and date
Florian Fainelli [Mon, 4 Jul 2016 21:36:09 +0000 (14:36 -0700)]
net: r6040: Bump version and date

Bump version to 0.28 and date to 4th of July 2016.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: r6040: Update my email
Florian Fainelli [Mon, 4 Jul 2016 21:36:08 +0000 (14:36 -0700)]
net: r6040: Update my email

Update my email address in the driver and MAINTAINERS file.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: r6040: Utilize napi_complete_done()
Florian Fainelli [Mon, 4 Jul 2016 21:36:07 +0000 (14:36 -0700)]
net: r6040: Utilize napi_complete_done()

We maintain how much work we did in NAPI context, so provide that with
napi_complete_done().

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: r6040: Utilize __napi_schedule_irqoff
Florian Fainelli [Mon, 4 Jul 2016 21:36:06 +0000 (14:36 -0700)]
net: r6040: Utilize __napi_schedule_irqoff

We are already in hard IRQ context, so we can use
__napi_schedule_irqoff() to save a few operations.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: r6040: Check for skb->xmit_more
Florian Fainelli [Mon, 4 Jul 2016 21:36:05 +0000 (14:36 -0700)]
net: r6040: Check for skb->xmit_more

Kick the transmission only if this is the last SKB to transmit or the
queue is not already stopped.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: r6040: Reclaim transmitted buffers in NAPI
Florian Fainelli [Mon, 4 Jul 2016 21:36:04 +0000 (14:36 -0700)]
net: r6040: Reclaim transmitted buffers in NAPI

Instead of taking one interrupt per packet transmitted, re-use the same
NAPI context to free transmitted buffers. Since we are no longer in hard
IRQ context replace dev_kfree_skb_irq() by dev_kfree_skb().

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: r6040: Utilize skb_put_padto()
Florian Fainelli [Mon, 4 Jul 2016 21:36:03 +0000 (14:36 -0700)]
net: r6040: Utilize skb_put_padto()

Pad the SKB to the minimum length of ETH_ZLEN by using skb_put_padto()
and take this operation out of the critical section since there is no
need to check any HW resources before doing that.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: r6040: Increase statistics upon transmit completion
Florian Fainelli [Mon, 4 Jul 2016 21:36:02 +0000 (14:36 -0700)]
net: r6040: Increase statistics upon transmit completion

r6040_xmit() is increasing transmit statistics during transmission while
this may still fail, do this in r6040_tx() where we complete transmitted
buffers instead.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: r6040: Utilize phy_print_status
Florian Fainelli [Mon, 4 Jul 2016 21:36:01 +0000 (14:36 -0700)]
net: r6040: Utilize phy_print_status

Instead of open coding our own version utilize the library provided
function.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agodwc_eth_qos: remove unused including <linux/version.h>
Wei Yongjun [Mon, 4 Jul 2016 15:24:50 +0000 (15:24 +0000)]
dwc_eth_qos: remove unused including <linux/version.h>

Remove including <linux/version.h> that don't need it.

Signed-off-by: Wei Yongjun <weiyj.lk@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'mlx5-ntuple-steering'
David S. Miller [Tue, 5 Jul 2016 07:06:03 +0000 (00:06 -0700)]
Merge branch 'mlx5-ntuple-steering'

Saeed Mahameed says:

====================
Mellanox 100G mlx5 ethtool ntuple steering

This series adds Ethernet ethtool ntuple steering 'ethtool -N|U' and exposes two more
counter sets to Ethtool statistics, RDMA vport and global flow control statistics.

We start from three refactoring patches of the flow steering infrastructure
    - mlx5_add_flow_rule will now receive mlx5 flow spec to simplify and reduce
      number of parameters
    - All low level steering objects are now wrapped in mlx5_flow_steering structure
      for better encapsulation
    - Flow steering object will now be removed properly and generically rather than
      traversing on a well-known steering tree objects

Patch#4 adds the infrastructure and the data structures needed for the ethtool ntuple
steering, all implemented in a new file 'en_fs_ethtool.c'.  Add the support for set_rxnfc
ethtool callback to add/remove/replace a flow spec of ethter type L2.

Patch#5 adds the support for L3/L4 flow specs and a higher priority in favor for L3/L4
rules when interleaving with L2 rules.

Patch#6 adds the support for get_rxnfc ethtool callback.

Patch#7,8 adds RDMA vport and global flow control statistics.

Applied on top: 8186f6e382d8 ('net-next: mediatek: fix compile error inside mtk_poll_controller()')
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5e: Expose flow control counters to ethtool
Gal Pressman [Mon, 4 Jul 2016 14:23:12 +0000 (17:23 +0300)]
net/mlx5e: Expose flow control counters to ethtool

Just like per prio counters, the global flow counters are queried from
per priority counters register.
Global flow control counters are stored in priority 0 PFC counters.

Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5e: Expose RDMA VPort counters to ethtool
Gal Pressman [Mon, 4 Jul 2016 14:23:11 +0000 (17:23 +0300)]
net/mlx5e: Expose RDMA VPort counters to ethtool

Add the needed descriptors to expose RoCE RDMA counters.

Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5e: Add support to get ethtool flow rules
Maor Gottlieb [Mon, 4 Jul 2016 14:23:10 +0000 (17:23 +0300)]
net/mlx5e: Add support to get ethtool flow rules

Enhance the existing get_rxnfc callback:
1. Get flow rule of specific ID.
2. Get all flow rules.
3. Get number of rules.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5e: Support l3/l4 flow type specs in ethtool flow steering
Maor Gottlieb [Mon, 4 Jul 2016 14:23:09 +0000 (17:23 +0300)]
net/mlx5e: Support l3/l4 flow type specs in ethtool flow steering

Add support to add flow steering rules with ethtool
of L3/L4 flow types (ip4/tcp4/udp4).
Those rules will be in higher priority than l2 flow rules, in order
to prefer more specific rules.

Mask is not supported for l3/l4 flow types.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5e: Add ethtool flow steering support
Maor Gottlieb [Mon, 4 Jul 2016 14:23:08 +0000 (17:23 +0300)]
net/mlx5e: Add ethtool flow steering support

Implement etrhtool set_rxnfc callback to support ethtool flow spec
direct steering. This patch adds only the support of ether flow type
spec. L3/L4 flow specs support will be added in downstream patches.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5: Properly remove all steering objects
Maor Gottlieb [Mon, 4 Jul 2016 14:23:07 +0000 (17:23 +0300)]
net/mlx5: Properly remove all steering objects

Instead of explicitly cleaning up the well known parts of the steering
tree, we use the generic tree structure to traverse for cleanup.
No functional changes.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5: Introduce mlx5_flow_steering structure
Maor Gottlieb [Mon, 4 Jul 2016 14:23:06 +0000 (17:23 +0300)]
net/mlx5: Introduce mlx5_flow_steering structure

Instead of having all steering private name spaces and
steering module fields flat in mlx5_core_priv, we wrap
them in mlx5_flow_steering for better modularity and
API exposure.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5: Refactor mlx5_add_flow_rule
Maor Gottlieb [Mon, 4 Jul 2016 14:23:05 +0000 (17:23 +0300)]
net/mlx5: Refactor mlx5_add_flow_rule

Reduce the set of arguments passed to mlx5_add_flow_rule
by introducing flow_spec structure.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet-next: mediatek: remove superfluous free_irq() call
John Crispin [Mon, 4 Jul 2016 13:37:10 +0000 (15:37 +0200)]
net-next: mediatek: remove superfluous free_irq() call

Commit 8067302973a1 ("net-next: mediatek: add support for IRQ grouping")
adds handling for irq 1 and 2 to the uninit function but did not remove
irq 0 which is not used since irq grouping was introduced. Fix this by
removing the superfluous call to free_irq().

Reported-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: John Crispin <john@phrozen.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: lpc_eth: Remove unused 'pldat' variable
Fabio Estevam [Mon, 4 Jul 2016 13:13:21 +0000 (10:13 -0300)]
net: lpc_eth: Remove unused 'pldat' variable

Since commit f786f3564c4f02d5026 ("net: ethernet: lpc_eth: use phydev
from struct net_device") the 'pldat' variable became unused, so
just remove it.

Reported-by: Olof's autobuilder <build@lixom.net>
Signed-off-by: Fabio Estevam <festevam@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'batadv-next-for-davem-20160704' of git://git.open-mesh.org/linux-merge
David S. Miller [Tue, 5 Jul 2016 06:33:59 +0000 (23:33 -0700)]
Merge tag 'batadv-next-for-davem-20160704' of git://git.open-mesh.org/linux-merge

Simon Wunderlich says:

====================
This feature patchset includes the following changes:

 - Cleanup work by Markus Pargmann and Sven Eckelmann (six patches)

 - Initial Netlink support by Matthias Schiffer (two patches)

 - Throughput Meter implementation by Antonio Quartulli, a kernel-space
   traffic generator to estimate link speeds. This feature is useful on
   low-end WiFi APs where running iperf or netperf from userspace
   gives wrong results due to heavy userspace/kernelspace overhead.
   (two patches)

 - API clean-up work by Antonio Quartulli (one patch)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqeth: delete napi struct when removing a qeth device
Ursula Braun [Mon, 4 Jul 2016 12:07:16 +0000 (14:07 +0200)]
qeth: delete napi struct when removing a qeth device

A qeth_card contains a napi_struct linked to the net_device during
device probing. This struct must be deleted when removing the qeth
device, otherwise Panic on oops can occur when qeth devices are
repeatedly removed and added.

Fixes: a1c3ed4c9ca ("qeth: NAPI support for l2 and l3 discipline")
Cc: stable@vger.kernel.org # v2.6.37+
Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Tested-by: Alexander Klein <ALKL@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'mlxsw-fib-offload'
David S. Miller [Tue, 5 Jul 2016 01:25:16 +0000 (18:25 -0700)]
Merge branch 'mlxsw-fib-offload'

Jiri Pirko says:

====================
mlxsw: Implement basic FIB offload and router interfaces

Introduce LPM trees management including virtual router management for HW.
Implement basic FIB offloading using switchdev FIB objects. For now only support
local routes and direct routes (next-hop support will be introduced in
a follow-up patchset).

Introduce router interfaces in patches 10-14.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Enable L3 interfaces on top of bridge devices
Ido Schimmel [Mon, 4 Jul 2016 06:23:17 +0000 (08:23 +0200)]
mlxsw: spectrum: Enable L3 interfaces on top of bridge devices

As with the previously introduced L3 interfaces, listen to 'inetaddr'
notifications sent for bridges devices configured on top of the port
netdevs and create / destroy router interfaces (RIFs) accordingly.
This also includes VLAN devices configured on top of the VLAN-aware
bridge.

The RIFs will be destroyed either when the last IP address is removed or
when the underlying FID is is destroyed.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Configure FIDs based on bridge events
Ido Schimmel [Mon, 4 Jul 2016 06:23:16 +0000 (08:23 +0200)]
mlxsw: spectrum: Configure FIDs based on bridge events

Before introducing support for L3 interfaces on top of the VLAN-aware
bridge we need to add some missing infrastructure.

Such an interface can either be the bridge device itself or a VLAN
device on top of it. In the first case the router interface (RIF) is
associated with FID 1, which is created whenever the first port netdev
joins the bridge. We currently assume the default PVID is 1 and that
it's already created, as it seems reasonable. This can be extended in
the future.

However, in the second case it's entirely possible we've yet to create a
matching FID. This can happen if the VLAN device was configured before
making any bridge port member in the VLAN.

Prevent such ordering problems by using the VLAN device's CHANGEUPPER
event to configure the FID. Make the VLAN device hold a reference to the
FID and prevent it from being destroyed even if none of the port netdevs
is using it.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Unsplit the vFID range
Ido Schimmel [Mon, 4 Jul 2016 06:23:15 +0000 (08:23 +0200)]
mlxsw: spectrum: Unsplit the vFID range

Previous commit deprecated the vFIDs used to get traffic to the CPU
('port_vfids'). Thus, we now use the vFIDs as god intended and the
artificial split is no longer needed.

Rename functions and variables to reflect that.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Introduce support for router interfaces
Ido Schimmel [Mon, 4 Jul 2016 06:23:14 +0000 (08:23 +0200)]
mlxsw: spectrum: Introduce support for router interfaces

Up until now we only supported bridged interfaces. Packets ingressing
through the switch ports were either classified to FIDs (in the case of
the VLAN-aware bridge) or vFIDs (in the case of VLAN-unaware bridges).
The packets were then forwarded according to the FDB. Routing was done
entirely in slowpath, by splitting the vFID range in two and using the
lower 0.5K vFIDs as dummy bridges that simply flooded all incoming
traffic to the CPU.

Instead, allow packets to be routed in the device by creating router
interfaces (RIFs) that will direct them to the router block.
Specifically, the RIFs introduced here are Sub-port RIFs used for VLAN
devices and port netdevs. Packets ingressing from the {Port / LAG ID, VID}
with which the RIF was programmed with will be assigned to a special
kind of FIDs called rFIDs and from there directed to the router.

Create a RIF whenever the first IPv4 address was programmed on a VLAN /
LAG / port netdev. Destroy it upon removal of the last IPv4 address.
Receive these notifications by registering for the 'inetaddr'
notification chain. A non-zero (10) priority is used for the
notification block, so that RIFs will be created before routes are
offloaded via FIB code.

Note that another trigger for RIF destruction are CHANGEUPPER
notifications causing the underlying FID's reference count to go down to
zero. This can happen, for example, when a VLAN netdev with an IP address
is put under bridge. While this configuration doesn't make sense it does
cause the device and the kernel to get out of sync when the netdev is
unbridged. We intend to address this in the future, hopefully in current
cycle.

Finally, Remove the lower 0.5K vFIDs, as they are deprecated by the RIFs,
which will trap packets according to their DIP.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Edit RIF properties based on netdev events
Ido Schimmel [Mon, 4 Jul 2016 06:23:13 +0000 (08:23 +0200)]
mlxsw: spectrum: Edit RIF properties based on netdev events

We are just about to introduce router interfaces (RIFs), but before that
we need to be able update the device with the correct RIF attributes
whenever they change for the netdev the RIF is backing. Two such
attributes are MTU and MAC.

The MAC is used both to set the source MAC of packets egressing from the
RIF and also to program an FDB rule that will direct packets to the
router block.

Use the existing netdevice notification block and respond to CHANGEADDR
and CHANGEMTU accordingly. Store both attributes in the RIF struct
in case we need to revert to old attributes following a failed update.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Add couple of lower device helper functions
Jiri Pirko [Mon, 4 Jul 2016 06:23:12 +0000 (08:23 +0200)]
mlxsw: spectrum: Add couple of lower device helper functions

Add functions that iterate over lower devices and find port device.
As a dependency add netdev_for_each_all_lower_dev and
netdev_for_each_all_lower_dev_rcu macro with
netdev_all_lower_get_next and netdev_all_lower_get_next_rcu shelpers.

Also, add functions to return mlxsw struct according to lower device
found and mlxsw_port struct with a reference to lower device.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum_router: Implement fib4 add/del switchdev obj ops
Jiri Pirko [Mon, 4 Jul 2016 06:23:11 +0000 (08:23 +0200)]
mlxsw: spectrum_router: Implement fib4 add/del switchdev obj ops

Implement ipv4 FIB entries addition and removal. Initially, we support
local and broadcast routes using "ip2me" trap action.
Also, unicast routes without nexthop are supported using "local" action.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: reg: Add Router Algorithmic LPM Unicast Entry Register definition
Jiri Pirko [Mon, 4 Jul 2016 06:23:10 +0000 (08:23 +0200)]
mlxsw: reg: Add Router Algorithmic LPM Unicast Entry Register definition

Serves for adding, updating and removing fib entries.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum_router: Add virtual router management
Jiri Pirko [Mon, 4 Jul 2016 06:23:09 +0000 (08:23 +0200)]
mlxsw: spectrum_router: Add virtual router management

Virtual router is a construct used inside HW. In this implementation
we map kernel tables to virtual routers one to one. Introduce management
logic to create virtual routers when needed and destroy in case they are
no longer in use. According to that, call into LPM tree management.
Each virtual router is always bound to one LPM tree.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>