linux-2.6-microblaze.git
6 years agonetfilter: xt_limit: Spelling s/maxmum/maximum/
Geert Uytterhoeven [Fri, 2 Mar 2018 13:59:54 +0000 (14:59 +0100)]
netfilter: xt_limit: Spelling s/maxmum/maximum/

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: make xt_rateest hash table per net
Cong Wang [Fri, 2 Mar 2018 02:58:38 +0000 (18:58 -0800)]
netfilter: make xt_rateest hash table per net

As suggested by Eric, we need to make the xt_rateest
hash table and its lock per netns to reduce lock
contentions.

Cc: Florian Westphal <fw@strlen.de>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: x_tables: ensure last rule in base chain matches underflow/policy
Florian Westphal [Tue, 27 Feb 2018 18:42:37 +0000 (19:42 +0100)]
netfilter: x_tables: ensure last rule in base chain matches underflow/policy

Harmless from kernel point of view, but again iptables assumes that
this is true when decoding ruleset coming from kernel.

If a (syzkaller generated) ruleset doesn't have the underflow/policy
stored as the last rule in the base chain, then iptables will abort()
because it doesn't find the chain policy.

libiptc assumes that the policy is the last rule in the basechain, which
is only true for iptables-generated rulesets.

Unfortunately this needs code duplication -- the functions need the
struct layout of the rule head, but that is different for
ip/ip6/arptables.

NB: pr_warn could be pr_debug but in case this break rulesets somehow its
useful to know why blob was rejected.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: x_tables: make sure compat af mutex is held
Florian Westphal [Tue, 27 Feb 2018 18:42:36 +0000 (19:42 +0100)]
netfilter: x_tables: make sure compat af mutex is held

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: compat: reject huge allocation requests
Florian Westphal [Tue, 27 Feb 2018 18:42:35 +0000 (19:42 +0100)]
netfilter: compat: reject huge allocation requests

no need to bother even trying to allocating huge compat offset arrays,
such ruleset is rejected later on anyway becaus we refuse to allocate
overly large rule blobs.

However, compat translation happens before blob allocation, so we should
add a check there too.

This is supposed to help with fuzzing by avoiding oom-killer.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: compat: prepare xt_compat_init_offsets to return errors
Florian Westphal [Tue, 27 Feb 2018 18:42:34 +0000 (19:42 +0100)]
netfilter: compat: prepare xt_compat_init_offsets to return errors

should have no impact, function still always returns 0.
This patch is only to ease review.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: x_tables: add counters allocation wrapper
Florian Westphal [Tue, 27 Feb 2018 18:42:33 +0000 (19:42 +0100)]
netfilter: x_tables: add counters allocation wrapper

allows to have size checks in a single spot.
This is supposed to reduce oom situations when fuzz-testing xtables.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: x_tables: limit allocation requests for blob rule heads
Florian Westphal [Tue, 27 Feb 2018 18:42:32 +0000 (19:42 +0100)]
netfilter: x_tables: limit allocation requests for blob rule heads

This is a very conservative limit (134217728 rules), but good
enough to not trigger frequent oom from syzkaller.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: x_tables: cap allocations at 512 mbyte
Florian Westphal [Tue, 27 Feb 2018 18:42:31 +0000 (19:42 +0100)]
netfilter: x_tables: cap allocations at 512 mbyte

Arbitrary limit, however, this still allows huge rulesets
(> 1 million rules).  This helps with automated fuzzer as it prevents
oom-killer invocation.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: x_tables: enforce unique and ascending entry points
Florian Westphal [Tue, 27 Feb 2018 18:42:30 +0000 (19:42 +0100)]
netfilter: x_tables: enforce unique and ascending entry points

Harmless from kernel point of view, but iptables assumes that this is
true when decoding a ruleset.

iptables walks the dumped blob from kernel, and, for each entry that
creates a new chain it prints out rule/chain information.
Base chains (hook entry points) are thus only shown when they appear
in the rule blob.  One base chain that is referenced multiple times
in hook blob is then only printed once.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: x_tables: move hook entry checks into core
Florian Westphal [Tue, 27 Feb 2018 18:42:29 +0000 (19:42 +0100)]
netfilter: x_tables: move hook entry checks into core

Allow followup patch to change on location instead of three.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: x_tables: check error target size too
Florian Westphal [Tue, 27 Feb 2018 18:42:28 +0000 (19:42 +0100)]
netfilter: x_tables: check error target size too

Check that userspace ERROR target (custom user-defined chains) match
expected format, and the chain name is null terminated.

This is irrelevant for kernel, but iptables itself relies on sane input
when it dumps rules from kernel.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: x_tables: check standard verdicts in core
Florian Westphal [Tue, 27 Feb 2018 18:42:27 +0000 (19:42 +0100)]
netfilter: x_tables: check standard verdicts in core

Userspace must provide a valid verdict to the standard target.

The verdict can be either a jump (signed int > 0), or a return code.

Allowed return codes are either RETURN (pop from stack), NF_ACCEPT, DROP
and QUEUE (latter is allowed for legacy reasons).

Jump offsets (verdict > 0) are checked in more detail later on when
loop-detection is performed.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: unlock xt_table earlier in __do_replace
Xin Long [Fri, 16 Feb 2018 10:04:56 +0000 (18:04 +0800)]
netfilter: unlock xt_table earlier in __do_replace

Now it's doing cleanup_entry for oldinfo under the xt_table lock,
but it's not really necessary. After the replacement job is done
in xt_replace_table, oldinfo is not used elsewhere any more, and
it can be freed without xt_table lock safely.

The important thing is that rtnl_lock is called in some xt_target
destroy, which means rtnl_lock, a big lock is used in xt_table
lock, a smaller one. It usually could be the reason why a dead
lock may happen.

Besides, all xt_target/match checkentry is called out of xt_table
lock. It's better also to move all cleanup_entry calling out of
xt_table lock, just as do_replace_finish does for ebtables.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: ipt_ah: return boolean instead of integer
Gustavo A. R. Silva [Tue, 13 Feb 2018 14:25:57 +0000 (08:25 -0600)]
netfilter: ipt_ah: return boolean instead of integer

Return statements in functions returning bool should use
true/false instead of 1/0.

This issue was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <garsilva@embeddedor.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: nf_conntrack_broadcast: remove useless parameter
Taehee Yoo [Sun, 11 Feb 2018 14:28:18 +0000 (23:28 +0900)]
netfilter: nf_conntrack_broadcast: remove useless parameter

parameter protoff in nf_conntrack_broadcast_help is not used anywhere.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: xt_cluster: get rid of xt_cluster_ipv6_is_multicast
Taehee Yoo [Sun, 11 Feb 2018 13:57:29 +0000 (22:57 +0900)]
netfilter: xt_cluster: get rid of xt_cluster_ipv6_is_multicast

If use the ipv6_addr_is_multicast instead of xt_cluster_ipv6_is_multicast,
then we can reduce code size.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: nfnetlink_acct: remove useless parameter
Taehee Yoo [Sun, 11 Feb 2018 10:17:20 +0000 (19:17 +0900)]
netfilter: nfnetlink_acct: remove useless parameter

parameter skb in nfnl_acct_overquota is not used anywhere.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: nf_tables: nf_tables_obj_lookup_byhandle() can be static
kbuild test robot [Fri, 19 Jan 2018 20:27:58 +0000 (04:27 +0800)]
netfilter: nf_tables: nf_tables_obj_lookup_byhandle() can be static

Fixes: 3ecbfd65f50e ("netfilter: nf_tables: allocate handle and delete objects via handle")
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agoMerge branch 'mvpp2-jumbo-frames-support'
David S. Miller [Mon, 5 Mar 2018 17:55:55 +0000 (12:55 -0500)]
Merge branch 'mvpp2-jumbo-frames-support'

Antoine Tenart says:

====================
net: mvpp2: jumbo frames support

This series enable jumbo frames support in the Marvell PPv2 driver. The
first 2 patches rework the buffer management, then two patches prepare for
the final patch which adds the jumbo frames support into the driver.

This is based on top of net-next, and was tested on a mcbin.

Thanks!
Antoine

Since v1:
  - Improved the Tx FIFO initialization comment.
  - Improved the pool sanity check in mvpp2_bm_pool_use().
  - Fixed pool related comments.
  - Cosmetic fixes (used BIT() whenever possible).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: mvpp2: jumbo frames support
Stefan Chulski [Mon, 5 Mar 2018 14:16:54 +0000 (15:16 +0100)]
net: mvpp2: jumbo frames support

This patch adds the support for jumbo frames in the Marvell PPv2 driver.
A third buffer pool is added with 10KB buffers, which is used if the MTU
is higher than 1518B for packets larger than 1518B. Please note only the
port 0 supports hardware checksum offload due to the Tx FIFO size
limitation.

Signed-off-by: Stefan Chulski <stefanc@marvell.com>
[Antoine: cosmetic cleanup, commit message]
Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: mvpp2: enable UDP/TCP checksum over IPv6
Antoine Tenart [Mon, 5 Mar 2018 14:16:53 +0000 (15:16 +0100)]
net: mvpp2: enable UDP/TCP checksum over IPv6

This patch adds the NETIF_F_IPV6_CSUM to the driver's features to enable
UDP/TCP checksum over IPv6. No extra configuration of the engine is
needed on top of the IPv4 counterpart, which already is in the features
list (NETIF_F_IP_CSUM).

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: mvpp2: use a data size of 10kB for Tx FIFO on port 0
Yan Markman [Mon, 5 Mar 2018 14:16:52 +0000 (15:16 +0100)]
net: mvpp2: use a data size of 10kB for Tx FIFO on port 0

This patch sets the Tx FIFO data size on port 0 to 10kB. This prepares
the PPv2 driver for the Jumbo frame support addition as the hardware
will need big enough Tx FIFO buffers when dealing with frames going
through an interface with an MTU of 9000.

Signed-off-by: Yan Markman <ymarkman@marvell.com>
[Antoine: commit message, small reworks.]
Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: mvpp2: update the BM buffer free/destroy logic
Stefan Chulski [Mon, 5 Mar 2018 14:16:51 +0000 (15:16 +0100)]
net: mvpp2: update the BM buffer free/destroy logic

The buffer free routine is updated to release only given a number of
buffers, and the destroy routine now checks the actual number of buffers
in the (BPPI and BPPE) HW counters before draining the pools. This
change helps getting jumbo frames support.

Signed-off-by: Stefan Chulski <stefanc@marvell.com>
[Antoine: cosmetic cleanup, commit message]
Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: mvpp2: use the same buffer pool for all ports
Stefan Chulski [Mon, 5 Mar 2018 14:16:50 +0000 (15:16 +0100)]
net: mvpp2: use the same buffer pool for all ports

This patch configures the buffer manager long pool for all ports part of
the same CP. Long pool separation between ports is redundant since there
are no performance improvement when different pools are used.

Signed-off-by: Stefan Chulski <stefanc@marvell.com>
[Antoine: cosmetic cleanup, commit message]
Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: core: dst: Add kernel-doc for 'net' parameter
Jonathan Neuschäfer [Sun, 4 Mar 2018 02:29:53 +0000 (03:29 +0100)]
net: core: dst: Add kernel-doc for 'net' parameter

This fixes the following kernel-doc warning:

./include/net/dst.h:366: warning: Function parameter or member 'net' not described in 'skb_tunnel_rx'

Fixes: ea23192e8e57 ("tunnels: harmonize cleanup done on skb on rx path")
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: core: dst_cache_set_ip6: Rename 'addr' parameter to 'saddr' for consistency
Jonathan Neuschäfer [Sun, 4 Mar 2018 02:29:52 +0000 (03:29 +0100)]
net: core: dst_cache_set_ip6: Rename 'addr' parameter to 'saddr' for consistency

The other dst_cache_{get,set}_ip{4,6} functions, and the doc comment for
dst_cache_set_ip6 use 'saddr' for their source address parameter. Rename
the parameter to increase consistency.

This fixes the following kernel-doc warnings:

./include/net/dst_cache.h:58: warning: Function parameter or member 'addr' not described in 'dst_cache_set_ip6'
./include/net/dst_cache.h:58: warning: Excess function parameter 'saddr' description in 'dst_cache_set_ip6'

Fixes: 911362c70df5 ("net: add dst_cache support")
Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: core: dst_cache: Fix a typo in a comment
Jonathan Neuschäfer [Sun, 4 Mar 2018 02:29:51 +0000 (03:29 +0100)]
net: core: dst_cache: Fix a typo in a comment

Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'convert-pernet_operations-part4'
David S. Miller [Mon, 5 Mar 2018 15:48:29 +0000 (10:48 -0500)]
Merge branch 'convert-pernet_operations-part4'

Kirill Tkhai says:

====================
Converting pernet_operations (part #4)

this series continues to review and to convert pernet_operations
to make them possible to be executed in parallel for several
net namespaces in the same time. The patches touch mostly netfilter,
also there are small number of changes in other places.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert proto_gre_net_ops
Kirill Tkhai [Mon, 5 Mar 2018 11:32:23 +0000 (14:32 +0300)]
net: Convert proto_gre_net_ops

These pernet_operations register and unregister sysctl.
nf_conntrack_l4proto_gre4->init_net is simple memory
initializer. Also, exit method removes gre keymap_list,
which is per-net. This looks safe to be executed
in parallel with other pernet_operations.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert ctnetlink_net_ops
Kirill Tkhai [Mon, 5 Mar 2018 11:32:15 +0000 (14:32 +0300)]
net: Convert ctnetlink_net_ops

These pernet_operations register and unregister
two conntrack notifiers, and they seem to be safe
to be executed in parallel.

General/not related to async pernet_operations JFI:
ctnetlink_net_exit_batch() actions are grouped in batch,
and this could look like there is synchronize_rcu()
is forgotten. But there is synchronize_rcu() on module
exit patch (in ctnetlink_exit()), so this batch may
be reworked as simple .exit method.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert nf_conntrack_net_ops
Kirill Tkhai [Mon, 5 Mar 2018 11:32:06 +0000 (14:32 +0300)]
net: Convert nf_conntrack_net_ops

These pernet_operations register and unregister sysctl and /proc
entries. Exit batch method also waits till all per-net conntracks
are dead. Thus, they are safe to be marked as async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert ip_set_net_ops
Kirill Tkhai [Mon, 5 Mar 2018 11:31:55 +0000 (14:31 +0300)]
net: Convert ip_set_net_ops

These pernet_operations initialize and destroy
net_generic(net, ip_set_net_id)-related data.
Since ip_set is under CONFIG_IP_SET, it's easy
to watch drivers, which depend on this config.
All of them are in net/netfilter/ipset directory,
except of net/netfilter/xt_set.c. There are no
more drivers, which use ip_set, and all of
the above don't register another pernet_operations.
Also, there are is no indirect users, as header
file include/linux/netfilter/ipset/ip_set.h does
not define indirect users by something like this:

#ifdef CONFIG_IP_SET
extern func(void);
#else
static inline func(void);
#endif

So, there are no more pernet operations, dereferencing
net_generic(net, ip_set_net_id).

ip_set_net_ops are OK to be executed in parallel
for several net, so we mark them as async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert fou_net_ops
Kirill Tkhai [Mon, 5 Mar 2018 11:31:47 +0000 (14:31 +0300)]
net: Convert fou_net_ops

These pernet_operations initialize and destroy
pernet net_generic(net, fou_net_id) list.
The rest of net_generic(net, fou_net_id) accesses
may happen after netlink message, and in-tree
pernet_operations do not send FOU_GENL_NAME messages.
So, these pernet_operations are safe to be marked
as async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert dccp_v6_ops
Kirill Tkhai [Mon, 5 Mar 2018 11:31:37 +0000 (14:31 +0300)]
net: Convert dccp_v6_ops

These pernet_operations looks similar to dccp_v4_ops,
and they are also safe to be marked as async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert dccp_v4_ops
Kirill Tkhai [Mon, 5 Mar 2018 11:31:28 +0000 (14:31 +0300)]
net: Convert dccp_v4_ops

These pernet_operations create and destroy net::dccp::v4_ctl_sk.
It looks like another pernet_operations don't want to send
dccp packets to dying or creating net. Batch method similar
to ipv4/ipv6 sockets and it has to be safe to be executed
in parallel with anything else. So, we mark them as async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert cangw_pernet_ops
Kirill Tkhai [Mon, 5 Mar 2018 11:31:19 +0000 (14:31 +0300)]
net: Convert cangw_pernet_ops

These pernet_operations have a deal with cgw_list,
and the rest of accesses are made under rtnl_lock().
The only exception is cgw_dump_jobs(), which is
accessed under rcu_read_lock(). cgw_dump_jobs() is
called on netlink request, and it does not seem,
foreign pernet_operations want to send a net such
the messages. So, we mark them as async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert caif_net_ops
Kirill Tkhai [Mon, 5 Mar 2018 11:31:10 +0000 (14:31 +0300)]
net: Convert caif_net_ops

Init method just allocates memory for new cfg, and
assigns net_generic(net, caif_net_id). Despite there is
synchronize_rcu() on error path in cfcnfg_create(),
in real this function does not use global lists,
so it looks like this synchronize_rcu() is some legacy
inheritance. Exit method removes caif devices under
rtnl_lock().

There could be a problem, if someone from foreign net
pernet_operations dereference caif_net_id of this net.
It's dereferenced in get_cfcnfg() and caif_device_list().

get_cfcnfg() is used from netdevice notifiers, where
they are called under rtnl_lock(). The notifiers can't
be called from foreign nets pernet_operations. Also,
it's used from caif_disconnect_client() and from
caif_connect_client(). The both of the functions work
with caif socket, and there is the only possibility
to have a socket, when the net is dead. This may happen
only of the socket was created as kern using sk_alloc().
Grep by PF_CAIF shows we do not create kern caif sockets,
so get_cfcnfg() is safe.

caif_device_list() is used in netdevice notifiers and exit
method under rtnl lock. Also, from caif_get() used in
the netdev notifiers and in caif_flow_cb(). The last item
is skb destructor. Since there are no kernel caif sockets
nobody can send net a packet in parallel with init/exit,
so this is also safe.

So, these pernet_operations are safe to be async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert arp_tables_net_ops and ip6_tables_net_ops
Kirill Tkhai [Mon, 5 Mar 2018 11:31:00 +0000 (14:31 +0300)]
net: Convert arp_tables_net_ops and ip6_tables_net_ops

These pernet_operations call xt_proto_init() and xt_proto_fini(),
which just register and unregister /proc entries.
They are safe to be marked as async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert log pernet_operations
Kirill Tkhai [Mon, 5 Mar 2018 11:30:50 +0000 (14:30 +0300)]
net: Convert log pernet_operations

These pernet_operations use nf_log_set() and nf_log_unset()
in their methods:

nf_log_bridge_net_ops
nf_log_arp_net_ops
nf_log_ipv4_net_ops
nf_log_ipv6_net_ops
nf_log_netdev_net_ops

Nobody can send such a packet to a net before it's became
registered, nobody can send a packet after all netdevices
are unregistered. So, these pernet_operations are able
to be marked as async.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Convert broute_net_ops, frame_filter_net_ops and frame_nat_net_ops
Kirill Tkhai [Mon, 5 Mar 2018 11:30:41 +0000 (14:30 +0300)]
net: Convert broute_net_ops, frame_filter_net_ops and frame_nat_net_ops

These pernet_operations use ebt_register_table() and
ebt_unregister_table() to act on the tables, which
are used as argument in ebt_do_table(), called from
ebtables hooks.

Since there are no net-related bridge packets in-flight,
when the init and exit methods are called, these
pernet_operations are safe to be executed in parallel
with any other.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: forwarding: Add suppport to create veth interfaces
David Ahern [Mon, 5 Mar 2018 01:37:47 +0000 (17:37 -0800)]
selftests: forwarding: Add suppport to create veth interfaces

For tests using veth interfaces, the test infrastructure can create
the netdevs if they do not exist. Arguably this is a preferred approach
since the tests require p$N and p$(N+1) to be pairs.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ncsi: Add generic netlink family
Samuel Mendoza-Jonas [Mon, 5 Mar 2018 00:39:05 +0000 (11:39 +1100)]
net/ncsi: Add generic netlink family

Add a generic netlink family for NCSI. This supports three commands;
NCSI_CMD_PKG_INFO which returns information on packages and their
associated channels, NCSI_CMD_SET_INTERFACE which allows a specific
package or package/channel combination to be set as the preferred
choice, and NCSI_CMD_CLEAR_INTERFACE which clears any preferred setting.

Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: add ca_state stat in SCM_TIMESTAMPING_OPT_STATS
Priyaranjan Jha [Sun, 4 Mar 2018 18:38:36 +0000 (10:38 -0800)]
tcp: add ca_state stat in SCM_TIMESTAMPING_OPT_STATS

This patch adds TCP_NLA_CA_STATE stat into SCM_TIMESTAMPING_OPT_STATS.
It reports ca_state of socket, when timestamp is generated.

Signed-off-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: add send queue size stat in SCM_TIMESTAMPING_OPT_STATS
Priyaranjan Jha [Sun, 4 Mar 2018 18:38:35 +0000 (10:38 -0800)]
tcp: add send queue size stat in SCM_TIMESTAMPING_OPT_STATS

This patch adds TCP_NLA_SENDQ_SIZE stat into SCM_TIMESTAMPING_OPT_STATS.
It reports no. of bytes present in send queue, when timestamp is
generated.

Signed-off-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: Extend the tc action test for action mirror
Arkadi Sharshevsky [Sun, 4 Mar 2018 14:35:26 +0000 (16:35 +0200)]
selftests: Extend the tc action test for action mirror

Currently the tc action test is used only to test mirred redirect
action. This patch extends it for mirred mirror.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Arkadi Sharshevsky <arkadis@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Make RX-FCS and LRO mutually exclusive
Gal Pressman [Sun, 4 Mar 2018 12:12:04 +0000 (14:12 +0200)]
net: Make RX-FCS and LRO mutually exclusive

LRO and RX-FCS offloads cannot be enabled at the same time since it is
not clear what should happen to the FCS of each coalesced packet.
The FCS is not really part of the TCP payload, hence cannot be merged
into one big packet. On the other hand, providing one big LRO packet
with one FCS contradicts the RX-FCS feature goal.

Use the fix features mechanism in order to prevent intersection of the
features and drop LRO in case RX-FCS is requested.

Enabling RX-FCS while LRO is enabled will result in:
$ ethtool -K ens6 rx-fcs on
Actual changes:
large-receive-offload: off [requested on]
rx-fcs: on

Signed-off-by: Gal Pressman <galp@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoliquidio: Corrected Rx bytes counting
Intiyaz Basha [Sat, 3 Mar 2018 02:29:04 +0000 (18:29 -0800)]
liquidio: Corrected Rx bytes counting

Corrected stats mismatch between Host Tx and its peer Rx stats

Signed-off-by: Intiyaz Basha <intiyaz.basha@cavium.com>
Acked-by: Derek Chickles <derek.chickles@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet sched actions: corrected extack message
Roman Mashak [Sat, 3 Mar 2018 01:52:01 +0000 (20:52 -0500)]
net sched actions: corrected extack message

Signed-off-by: Roman Mashak <mrv@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'batadv-next-for-davem-20180302' of git://git.open-mesh.org/linux-merge
David S. Miller [Sun, 4 Mar 2018 23:45:39 +0000 (18:45 -0500)]
Merge tag 'batadv-next-for-davem-20180302' of git://git.open-mesh.org/linux-merge

Simon Wunderlich says:

====================
This cleanup patchset includes the following patches:

 - bump version strings, by Simon Wunderlich

 - bump copyright years, by Sven Eckelmann

 - fix macro indendation for checkpatch, by Sven Eckelmann

 - fix comparison operator for bool returning functions,
   by Sven Eckelmann

 - assume 2-byte packet alignments for all packet types,
   by Matthias Schiffer
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipvlan: forbid vlan devices on top of ipvlan
Paolo Abeni [Fri, 2 Mar 2018 15:03:32 +0000 (16:03 +0100)]
ipvlan: forbid vlan devices on top of ipvlan

Currently we allow the creation of 8021q devices on top of
ipvlan, but such devices are nonfunctional, as the underlying
ipvlan rx_hanlder hook can't match the relevant traffic.

Be explicit and forbid the creation of such nonfunctional devices.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agovirtio-net: re enable XDP_REDIRECT for mergeable buffer
Jason Wang [Fri, 2 Mar 2018 09:29:14 +0000 (17:29 +0800)]
virtio-net: re enable XDP_REDIRECT for mergeable buffer

XDP_REDIRECT support for mergeable buffer was removed since commit
7324f5399b06 ("virtio_net: disable XDP_REDIRECT in receive_mergeable()
case"). This is because we don't reserve enough tailroom for struct
skb_shared_info which breaks XDP assumption. So this patch fixes this
by reserving enough tailroom and using fixed size of rx buffer.

Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: rtnetlink: remove testns on test fail
Prashant Bhole [Fri, 2 Mar 2018 02:22:20 +0000 (11:22 +0900)]
selftests: rtnetlink: remove testns on test fail

This patch removes testns after test failure so that next test can
continue with clean ns

Signed-off-by: Prashant Bhole <bhole_prashant_q7@lab.ntt.co.jp>
Acked-by: William Tu <u9012063@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'gre-seq-collect_md'
David S. Miller [Sun, 4 Mar 2018 23:35:02 +0000 (18:35 -0500)]
Merge branch 'gre-seq-collect_md'

William Tu says:

====================
gre: add sequence number for collect md mode.

Currently GRE sequence number can only be used in native tunnel mode.
The first patch adds sequence number support for gre collect
metadata mode, and the second patch tests it using BPF.

RFC2890 defines GRE sequence number to be specific to the traffic
flow identified by the key.  However, this patch does not implement
per-key seqno.  The sequence number is shared in the same tunnel
device. That is, different tunnel keys using the same collect_md
tunnel share single sequence number.

A new BFP uapi tunnel flag 'BPF_F_SEQ_NUMBER' is added.
--
v1->v2:
  rename BPF_F_GRE_SEQ to BPF_F_SEQ_NUMBER suggested by Daniel
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosamples/bpf: add gre sequence number test.
William Tu [Thu, 1 Mar 2018 21:49:58 +0000 (13:49 -0800)]
samples/bpf: add gre sequence number test.

The patch adds tests for GRE sequence number
support for metadata mode tunnel.

Signed-off-by: William Tu <u9012063@gmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agogre: add sequence number for collect md mode.
William Tu [Thu, 1 Mar 2018 21:49:57 +0000 (13:49 -0800)]
gre: add sequence number for collect md mode.

Currently GRE sequence number can only be used in native
tunnel mode.  This patch adds sequence number support for
gre collect metadata mode.  RFC2890 defines GRE sequence
number to be specific to the traffic flow identified by the
key.  However, this patch does not implement per-key seqno.
The sequence number is shared in the same tunnel device.
That is, different tunnel keys using the same collect_md
tunnel share single sequence number.

Signed-off-by: William Tu <u9012063@gmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'enic-update'
David S. Miller [Sun, 4 Mar 2018 23:19:26 +0000 (18:19 -0500)]
Merge branch 'enic-update'

Govindarajulu Varadarajan says:

====================
enic update

This series adds support for IPv6 vxlan offload and UDP rss along with a
bug fix in filling the rq ring.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoenic: set IG desc cache flag in open
Govindarajulu Varadarajan [Thu, 1 Mar 2018 19:07:24 +0000 (11:07 -0800)]
enic: set IG desc cache flag in open

New adapter needs CMD_OPENF_IG_DESCCACHE flag to be set. If this flag is
not set, fw flushes the global IG desc cache. This flag is nop in older
adapter.

Also increment driver version

Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoenic: enable rq before updating rq descriptors
Govindarajulu Varadarajan [Thu, 1 Mar 2018 19:07:23 +0000 (11:07 -0800)]
enic: enable rq before updating rq descriptors

rq should be enabled before posting the buffers to rq desc. If not hw sees
stale value and casuses DMAR errors.

Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoenic: set UDP rss flag
Govindarajulu Varadarajan [Thu, 1 Mar 2018 19:07:22 +0000 (11:07 -0800)]
enic: set UDP rss flag

New hardware needs UDP flag set to enable UDP L4 rss hash. Add ethtool
get option to display supported rss flow hash.

Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoenic: Check if hw supports multi wq with vxlan offload
Govindarajulu Varadarajan [Thu, 1 Mar 2018 19:07:21 +0000 (11:07 -0800)]
enic: Check if hw supports multi wq with vxlan offload

Some adaptors do not support vxlan offload when multi wq is configured.

If hw supports multi wq, BIT(2) is set in a1.

Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoenic: Add vxlan offload support for IPv6 pkts
Govindarajulu Varadarajan [Thu, 1 Mar 2018 19:07:20 +0000 (11:07 -0800)]
enic: Add vxlan offload support for IPv6 pkts

New adaptors supports vxlan offload for inner IPv6 and outer IPv6 vxlan
pkts.

Fw sets BIT(0) & BIT(1) in a1 if hw supports ipv6 inner & outer pkt
offload.

Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoenic: Check inner ip proto for pseudo header csum
Govindarajulu Varadarajan [Thu, 1 Mar 2018 19:07:19 +0000 (11:07 -0800)]
enic: Check inner ip proto for pseudo header csum

To compute pseudo IP header csum, we need to check the inner header for
encap pkt, not outer IP header.

Also add pseudo csum for IPv6 inner pkt.

Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: amd8111e: remove redundant assignment to 'tx_index'
Colin Ian King [Thu, 1 Mar 2018 16:42:40 +0000 (16:42 +0000)]
net: amd8111e: remove redundant assignment to 'tx_index'

The variable tx_index is being initialized with a value that is never
read and re-assigned a little later, hence the initialization is redundant
and can be removed.

Cleans up clang warning:
drivers/net/ethernet/amd/amd8111e.c:652:6: warning: Value stored to
'tx_index' during its initialization is never read

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: switch to device-managed functions in probe (part 2)
Andy Shevchenko [Thu, 1 Mar 2018 11:27:35 +0000 (13:27 +0200)]
r8169: switch to device-managed functions in probe (part 2)

This is a follow up to the commit

  4c45d24a759d ("r8169: switch to device-managed functions in probe")

to move towards managed resources even more.

Cc: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: Dereference MMIO address immediately before use
Andy Shevchenko [Thu, 1 Mar 2018 11:27:34 +0000 (13:27 +0200)]
r8169: Dereference MMIO address immediately before use

There is no need to dereference struct rtl8169_private to get mmio_addr
in almost every function in the driver.

Replace it by using pointer to struct rtl8169_private directly.

No functional change intended.

Next step might be a conversion of RTL_Wxx() / RTL_Rxx() macros
to inline functions for sake of type checking.

Cc: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: Fix spelling mistake: "advertisment"-> "advertisement"
Colin Ian King [Thu, 1 Mar 2018 10:23:03 +0000 (10:23 +0000)]
net: phy: Fix spelling mistake: "advertisment"-> "advertisement"

Trivial fix to spelling mistake in comments and error message text.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocxgb4vf: Forcefully link up virtual interfaces
Arjun Vynipadath [Thu, 1 Mar 2018 09:31:04 +0000 (15:01 +0530)]
cxgb4vf: Forcefully link up virtual interfaces

The Virtual Interfaces are connected to an internal switch on the chip
which allows VIs attached to the same port to talk to each other even
when the port link is down.  As a result, we generally want to always
report a VI's link as being "up".

Based on the original work by: Casey Leedom <leedom@chelsio.com>
Signed-off-by: Arjun Vynipadath <arjun@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'dsa-serdes-stats'
David S. Miller [Sun, 4 Mar 2018 18:34:19 +0000 (13:34 -0500)]
Merge branch 'dsa-serdes-stats'

Andrew Lunn says:

====================
Export SERDES stats via ethtool -S

The mv88e6352 family has a SERDES interface which can be used for
example to connect to SFF/SFP modules. This interface has a couple of
statistics counters. Add support for including these counters in the
output of ethtool -S.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: mv88e6xxx: Get mv88e6352 SERDES statistics
Andrew Lunn [Thu, 1 Mar 2018 01:02:31 +0000 (02:02 +0100)]
net: dsa: mv88e6xxx: Get mv88e6352 SERDES statistics

Add support for reading the SERDES statistics of the mv88e8352, using
the standard ethtool -S option. The SERDES interface can be mapped to
either port 4 or 5, so only return statistics on those ports, if the
SERDES interface is in use.

The counters are reset on read, so need to be accumulated. Add a per
port structure to hold the stats counters. The 6352 only has a single
SERDES interface and so only one port will using the newly added
array. However the 6390 family has as many SERDES interfaces as ports,
each with statistics counters. Also, PTP has a number of counters per
port which will also need accumulating.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: mv88e6xxx: Add helper to determining if port has SERDES
Andrew Lunn [Thu, 1 Mar 2018 01:02:30 +0000 (02:02 +0100)]
net: dsa: mv88e6xxx: Add helper to determining if port has SERDES

Refactor the existing code. This helper will be used for SERDES
statistics.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: mv88e6xxx: Allow the SERDES interfaces to have statistics
Andrew Lunn [Thu, 1 Mar 2018 01:02:29 +0000 (02:02 +0100)]
net: dsa: mv88e6xxx: Allow the SERDES interfaces to have statistics

When gettting the number of statistics, the strings and the actual
statistics, call the SERDES ops if implemented. This means the stats
code needs to return the number of strings/stats they have placed into
the data, so that the SERDES strings/stats can follow on.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: mv88e6xxx: Hold mutex while doing stats operations
Andrew Lunn [Thu, 1 Mar 2018 01:02:28 +0000 (02:02 +0100)]
net: dsa: mv88e6xxx: Hold mutex while doing stats operations

Until now, there has been no need to hold the reg mutex while getting
the count of statistics, or the strings, because the hardware was not
accessed. When adding support for SERDES statistics, it is necessary
to access the hardware, to determine if a port is using the SERDES
interface. So add mutex lock/unlocks.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodsa: Pass the port to get_sset_count()
Andrew Lunn [Thu, 1 Mar 2018 01:02:27 +0000 (02:02 +0100)]
dsa: Pass the port to get_sset_count()

By passing the port, we allow different ports to have different
statistics. This is useful since some ports have SERDES interfaces
with their own statistic counters.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotools: tc-testing: Add notap option
Brenda J. Butler [Wed, 28 Feb 2018 20:36:19 +0000 (15:36 -0500)]
tools: tc-testing: Add notap option

Add a command line arg to suppress tap output.  Handy in case
all the tap output is being supplied by the plugins.

Signed-off-by: Brenda J. Butler <bjb@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'net-ipv6-Add-support-for-path-selection-using-hash-of-5-tuple'
David S. Miller [Sun, 4 Mar 2018 18:04:24 +0000 (13:04 -0500)]
Merge branch 'net-ipv6-Add-support-for-path-selection-using-hash-of-5-tuple'

David Ahern says:

====================
net/ipv6: Add support for path selection using hash of 5-tuple

Hardware supports multipath selection using the standard L4 5-tuple
instead of just L3 and the flow label. In addition, some network
operators prefer IPv6 path selection to use the 5-tuple. To that end,
add support to IPv6 for multipath hash policy similar to
bf4e0a3db97eb ("net: ipv4: add support for ECMP hash policy choice").
The default is still L3 which covers source and destination addresses
along with flow label and IPv6 protocol. This gives users a choice in
hash algorithms if they believe L3 only and the IPv6 flow label are not
sufficient for their use case.

A separate sysctl is added for IPv6, allowing IPv4 and IPv6 to use
different algorithms if desired.

The first 3 patches modify the IPv4 variant so that at the end of the
patch set the ipv4 and ipv6 implementations are direct parallels.

Patch 4 refactors the existing rt6_multipath_hash in preparation for
adding the policy option.

Patch 5 renames the existing netevent to have IPv4 in the name so ipv4
changes can be distinguished from IPv6 if the netevent handler cares.

Patch 6 adds the skb as an argument through the FIB lookup functions
to the multipath selection. Needed for the forwarding case.

Patch 7 adds the L4 hash support.

Patch 8 adds the hook for the netevent to the spectrum driver to update
the ASIC.

Patch 9 removes no longer used code.

Patch 10 adds a testcase for IPv6 multipath with L4 hash.

v3
- comments from Ido:
  - removed fib_info arg in patch 1; left by mistake on rebase to net-next
  - removed __get_hash_from_flowi4 declaration
  - line wrap change to spectrum_router.c to maintain 80 chars

v2
- rebased to top of tree
- added refactor of fib_multipath_hash following recent change
- plumb skb through lookup functions to multipath selection
- fix sysctl setting; was missing the data set in ipv6_sysctl_net_init
- added test case

RFC to v1:
- rebase to top of net-next
- fix addr_type in hash_keys and removed flow label as noticed by Ido
- added a comment to cover letter about choice in algorithms based on
  use case per Or's comments
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: forwarding: Add multipath test for L4 hashing
David Ahern [Fri, 2 Mar 2018 16:32:21 +0000 (08:32 -0800)]
selftests: forwarding: Add multipath test for L4 hashing

Add IPv6 multipath test using L4 hashing. Created with inputs from
Ido Schimmel.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Tested-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Remove unused get_hash_from_flow functions
David Ahern [Fri, 2 Mar 2018 16:32:20 +0000 (08:32 -0800)]
net: Remove unused get_hash_from_flow functions

__get_hash_from_flowi6 is still used for flowlabels, but the IPv4
variant and the wrappers to both are not used. Remove them.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum_router: Add support for ipv6 hash policy update
David Ahern [Fri, 2 Mar 2018 16:32:19 +0000 (08:32 -0800)]
mlxsw: spectrum_router: Add support for ipv6 hash policy update

Similar to 28678f07f127d ("mlxsw: spectrum_router: Update multipath hash
parameters upon netevents") for IPv4, make sure the kernel and asic are
using the same hash algorithm for path selection.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Tested-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Add support for path selection using hash of 5-tuple
David Ahern [Fri, 2 Mar 2018 16:32:18 +0000 (08:32 -0800)]
net/ipv6: Add support for path selection using hash of 5-tuple

Some operators prefer IPv6 path selection to use a standard 5-tuple
hash rather than just an L3 hash with the flow the label. To that end
add support to IPv6 for multipath hash policy similar to bf4e0a3db97eb
("net: ipv4: add support for ECMP hash policy choice"). The default
is still L3 which covers source and destination addresses along with
flow label and IPv6 protocol.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Tested-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Pass skb to route lookup
David Ahern [Fri, 2 Mar 2018 16:32:17 +0000 (08:32 -0800)]
net/ipv6: Pass skb to route lookup

IPv6 does path selection for multipath routes deep in the lookup
functions. The next patch adds L4 hash option and needs the skb
for the forward path. To get the skb to the relevant FIB lookup
functions it needs to go through the fib rules layer, so add a
lookup_data argument to the fib_lookup_arg struct.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Rename NETEVENT_MULTIPATH_HASH_UPDATE
David Ahern [Fri, 2 Mar 2018 16:32:16 +0000 (08:32 -0800)]
net: Rename NETEVENT_MULTIPATH_HASH_UPDATE

Rename NETEVENT_MULTIPATH_HASH_UPDATE to
NETEVENT_IPV4_MPATH_HASH_UPDATE to denote it relates to a change
in the IPv4 hash policy.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Make rt6_multipath_hash similar to fib_multipath_hash
David Ahern [Fri, 2 Mar 2018 16:32:15 +0000 (08:32 -0800)]
net/ipv6: Make rt6_multipath_hash similar to fib_multipath_hash

Make rt6_multipath_hash more of a direct parallel to fib_multipath_hash
and reduce stack and overhead in the process: get_hash_from_flowi6 is
just a wrapper around __get_hash_from_flowi6 with another stack
allocation for flow_keys. Move setting the addresses, protocol and
label into rt6_multipath_hash and allow it to make the call to
flow_hash_from_keys.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv4: Simplify fib_multipath_hash with optional flow keys
David Ahern [Fri, 2 Mar 2018 16:32:14 +0000 (08:32 -0800)]
net/ipv4: Simplify fib_multipath_hash with optional flow keys

As of commit e37b1e978bec5 ("ipv6: route: dissect flow in input path if
fib rules need it") fib_multipath_hash takes an optional flow keys. If
non-NULL it means the skb has already been dissected. If not set, then
fib_multipath_hash needs to call skb_flow_dissect_flow_keys.

Simplify the logic by setting flkeys to the local stack variable keys.
Simplifies fib_multipath_hash by only have 1 set of instructions
setting hash_keys.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Align ip_multipath_l3_keys and ip6_multipath_l3_keys
David Ahern [Fri, 2 Mar 2018 16:32:13 +0000 (08:32 -0800)]
net: Align ip_multipath_l3_keys and ip6_multipath_l3_keys

Symmetry is good and allows easy comparison that ipv4 and ipv6 are
doing the same thing. To that end, change ip_multipath_l3_keys to
set addresses at the end after the icmp compares, and move the
initialization of ipv6 flow keys to rt6_multipath_hash.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv4: Pass net to fib_multipath_hash instead of fib_info
David Ahern [Fri, 2 Mar 2018 16:32:12 +0000 (08:32 -0800)]
net/ipv4: Pass net to fib_multipath_hash instead of fib_info

fib_multipath_hash only needs net struct to check a sysctl. Make it
clear by passing net instead of fib_info. In the end this allows
alignment between the ipv4 and ipv6 versions.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'sctp-clean-up-sctp_sendmsg'
David S. Miller [Sun, 4 Mar 2018 18:00:58 +0000 (13:00 -0500)]
Merge branch 'sctp-clean-up-sctp_sendmsg'

Xin Long says:

====================
sctp: clean up sctp_sendmsg

This cleanup mostly does three things:

 - extract some codes into functions to make sendmsg more readable.

 - tidy up some codes to avoid the unnecessary checks.

 - adjust some logic so that it will be easier to add the send flags
   and cmsgs features that I will post after this.

To make it easy to review and to check if the code is compatible with
before, this patchset is to do it step by step in 9 patches.

NOTE:
There will be a conflict when merging
Commit 2277c7cd75e3 ("sctp: Add LSM hooks") from selinux tree,
the solution is to:

1. remove all the lines in [B]:

    <<<<<<< HEAD
    [A]
    =======
    [B]
    >>>>>>> 2277c7c... sctp: Add LSM hooks

2. and apply the following diff-output:

diff --git a/net/sctp/socket.c b/net/sctp/socket.c
index 980621e..d6803c8 100644
--- a/net/sctp/socket.c
+++ b/net/sctp/socket.c
@@ -1686,6 +1686,7 @@ static int sctp_sendmsg_new_asoc(struct sock *sk, __u16 sflags,
  struct net *net = sock_net(sk);
  struct sctp_association *asoc;
  enum sctp_scope scope;
+ struct sctp_af *af;
  int err = -EINVAL;

  *tp = NULL;
@@ -1711,6 +1712,22 @@ static int sctp_sendmsg_new_asoc(struct sock *sk, __u16 sflags,

  scope = sctp_scope(daddr);

+ /* Label connection socket for first association 1-to-many
+  * style for client sequence socket()->sendmsg(). This
+  * needs to be done before sctp_assoc_add_peer() as that will
+  * set up the initial packet that needs to account for any
+  * security ip options (CIPSO/CALIPSO) added to the packet.
+  */
+ af = sctp_get_af_specific(daddr->sa.sa_family);
+ if (!af)
+ return -EINVAL;
+
+ err = security_sctp_bind_connect(sk, SCTP_SENDMSG_CONNECT,
+  (struct sockaddr *)daddr,
+  af->sockaddr_len);
+ if (err < 0)
+ return err;
+
  asoc = sctp_association_new(ep, sk, scope, GFP_KERNEL);
  if (!asoc)
  return -ENOMEM;
====================

Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: adjust some codes in a better order in sctp_sendmsg
Xin Long [Thu, 1 Mar 2018 15:05:18 +0000 (23:05 +0800)]
sctp: adjust some codes in a better order in sctp_sendmsg

sctp_sendmsg_new_asoc and SCTP_ADDR_OVER check is only necessary
when daddr is set, so move them up to if (daddr) statement.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: improve some variables in sctp_sendmsg
Xin Long [Thu, 1 Mar 2018 15:05:17 +0000 (23:05 +0800)]
sctp: improve some variables in sctp_sendmsg

This patch mostly is to:

  - rename sinfo_flags as sflags, to make the indents look better, and
    also keep consistent with other sctp_sendmsg_xx functions.

  - replace new_asoc with bool new, no need to define a pointer here,
    as if new_asoc is set, it must be asoc.

  - rename the 'out_nounlock:' as 'out', shorter and nicer.

  - remove associd, only one place is using it now, just use
    sinfo->sinfo_assoc_id directly.

  - remove 'cmsgs' initialization in sctp_sendmsg, as it will be done
    in sctp_sendmsg_parse.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the unnecessary transport looking up from sctp_sendmsg
Xin Long [Thu, 1 Mar 2018 15:05:16 +0000 (23:05 +0800)]
sctp: remove the unnecessary transport looking up from sctp_sendmsg

Now sctp_assoc_lookup_paddr can only be called only if daddr has
been set. But if daddr has been set, sctp_endpoint_lookup_assoc
would be done, where it could already have the transport.

So this unnecessary transport looking up should be removed, but
only reset transport as NULL when SCTP_ADDR_OVER is not set for
UDP type socket.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: factor out sctp_sendmsg_update_sinfo from sctp_sendmsg
Xin Long [Thu, 1 Mar 2018 15:05:15 +0000 (23:05 +0800)]
sctp: factor out sctp_sendmsg_update_sinfo from sctp_sendmsg

This patch is to move the codes for trying to get sinfo from
asoc into sctp_sendmsg_update_sinfo.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: factor out sctp_sendmsg_parse from sctp_sendmsg
Xin Long [Thu, 1 Mar 2018 15:05:14 +0000 (23:05 +0800)]
sctp: factor out sctp_sendmsg_parse from sctp_sendmsg

This patch is to move the codes for parsing msghdr and checking
sk into sctp_sendmsg_parse.

Note that different from before, 'sinfo' in sctp_sendmsg won't
be NULL any more. It gets the value either from cmsgs->srinfo,
cmsgs->sinfo or asoc. With it, the 'sinfo' and 'fill_sinfo_ttl'
check can be removed from sctp_sendmsg.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: factor out sctp_sendmsg_get_daddr from sctp_sendmsg
Xin Long [Thu, 1 Mar 2018 15:05:13 +0000 (23:05 +0800)]
sctp: factor out sctp_sendmsg_get_daddr from sctp_sendmsg

This patch is to move the codes for trying to get daddr from
msg->msg_name into sctp_sendmsg_get_daddr.

Note that after adding 'daddr', 'to' and 'msg_name' can be
deleted.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: factor out sctp_sendmsg_check_sflags from sctp_sendmsg
Xin Long [Thu, 1 Mar 2018 15:05:12 +0000 (23:05 +0800)]
sctp: factor out sctp_sendmsg_check_sflags from sctp_sendmsg

This patch is to move the codes for checking sinfo_flags on one asoc
after this asoc has been found into sctp_sendmsg_check_sflags.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: factor out sctp_sendmsg_new_asoc from sctp_sendmsg
Xin Long [Thu, 1 Mar 2018 15:05:11 +0000 (23:05 +0800)]
sctp: factor out sctp_sendmsg_new_asoc from sctp_sendmsg

This patch is to move the codes for creating a new asoc if
no asoc was found into sctp_sendmsg_new_asoc.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: factor out sctp_sendmsg_to_asoc from sctp_sendmsg
Xin Long [Thu, 1 Mar 2018 15:05:10 +0000 (23:05 +0800)]
sctp: factor out sctp_sendmsg_to_asoc from sctp_sendmsg

This patch is to move the codes for checking and sending on
one asoc after this asoc has been found or created into
sctp_sendmsg_to_asoc.

Note that 'err != -ESRCH' check is for the case that asoc is
freed when waiting for tx buffer in sctp_sendmsg_to_asoc.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
David S. Miller [Sat, 3 Mar 2018 02:53:11 +0000 (21:53 -0500)]
Merge git://git./linux/kernel/git/bpf/bpf-next

Daniel Borkmann says:

====================
pull-request: bpf-next 2018-03-03

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

The main changes are:

1) Extend bpftool to build up CFG information of eBPF programs and add an
   option to dump this in DOT format such that this can later be used with
   DOT graphic tools (xdot, graphviz, etc) to visualize it. Part of the
   analysis performed is sub-program detection and basic-block partitioning,
   from Jiong.

2) Multiple enhancements for bpftool's batch mode, more specifically the
   parser now understands comments (#), continuation lines (\), and arguments
   enclosed between quotes. Also, allow to read from stdin via '-' as input
   file, all from Quentin.

3) Improve BPF kselftests by i) unifying the rlimit handling into a helper
   that is then used by all tests, and ii) add support for testing tail calls
   to test_verifier plus add tests covering all corner cases. The latter is
   especially useful for testing JITs, from Daniel.

4) Remove x64 JIT's bpf_flush_icache() since flush_icache_range() is a noop
   on x64, from Daniel.

5) Fix one more occasion in BPF samples where we do not detach the BPF program
   from the cgroup after completion, from Prashant.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/usb/kalmia: use ARRAY_SIZE for various array sizing calculations
Colin Ian King [Fri, 2 Mar 2018 13:42:39 +0000 (13:42 +0000)]
net/usb/kalmia: use ARRAY_SIZE for various array sizing calculations

Use the ARRAY_SIZE macro on a couple of arrays to determine
size of the arrays. Also fix up alignment to clean up a checkpatch
warning. Improvement suggested by Coccinelle.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocxgb4: Add TP Congestion map entry for single-port
Ganesh Goudar [Fri, 2 Mar 2018 10:27:07 +0000 (15:57 +0530)]
cxgb4: Add TP Congestion map entry for single-port

Add TP Congestion Map entry for single-port T6 cards.

Signed-off-by: Casey Leedom <leedom@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'mac80211-next-for-davem-2018-03-02' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Fri, 2 Mar 2018 14:50:21 +0000 (09:50 -0500)]
Merge tag 'mac80211-next-for-davem-2018-03-02' of git://git./linux/kernel/git/jberg/mac80211-next

Johannes Berg says:

====================
Only a few new things:
 * hwsim net namespace stuff from Kirill Tkhai
 * A-MSDU support in fast-RX
 * 4-addr mode support in fast-RX
 * support for a spec quirk in Add-BA negotiation
====================

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