net: dsa: mv88e6352: Add support for ndo_fdb functions

Add support for manipulating switch fdb entries by pointing to the
ndo_fdb functions implemented for mv88e6xxxx.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6xxx: Add support for fdb_add, fdb_del, and fdb_getnext

No vlan support at this time.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: Add basic framework to support ndo_fdb functions

Provide callbacks for ndo_fdb_add, ndo_fdb_del, and ndo_fdb_dump.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6352: Add support for hardware bridging

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6xxx: Add Hardware bridging support

Bridge support is similar for all chips supported by the mv88e6xxx code,
so add the code there.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6171: Use common port configuration

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6123_61_65: Use common port configuration

This will simplify adding offloaded bridge support later on.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6352: Use common port initialization code

This prepares the driver for hardware bridging.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6xxx: Split mv88e6xxx_reg_read and mv88e6xxx_reg_write

Split mv88e6xxx_reg_read and mv88e6xxx_reg_write into two functions each,
one to acquire smi_mutex and one to get struct mii_bus *bus from
struct dsa_switch *ds and to call the actual read/write function.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6xxx: Disable Message Port bit for CPU port

Datasheet says that the Message Port bit should not be set for the CPU port.
Having it set causes DSA tagged packets to be sent to the CPU port roughly
every 30 seconds. Those packets are the same as real packets forwarded between
switch ports if the switch is configured for switching between multiple ports.
The packets are then bridged by the software bridge, resulting in duplicated
packets on the network.

Reported-by: Andrew Lunn <andrew@lunn.ch>
Cc: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6xxx: Provide function for common port initialization

Provide mv88e6xxx_setup_port_common() for common port initialization.
Currently only write Port 1 Control and VLAN configuration since
this will be needed for hardware bridging. More can be added later
if desired/needed.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6xxx: Factor out common initialization code

Code used and needed in mv886xxx.c should be initialized there as well,
so factor it out from the individual initialization files.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

hv_netvsc: remove vmbus_are_subchannels_present() in rndis_filter_device_add()

The vmbus_are_subchannels_present() also involves opening the channels, which
may be too early at this point. Checking for subchannels is not necessary here.
So this patch removes it. Subchannels will be opened when offer messages arrive.

Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: smc91x: make use of 4th parameter to devm_gpiod_get_index

Since 39b2bbe3d715 (gpio: add flags argument to gpiod_get*() functions)
which appeared in v3.17-rc1, the gpiod_get* functions take an additional
parameter that allows to specify direction and initial value for output.
Simplify accordingly.

Moreover use devm_gpiod_get_index_optional for still simpler handling.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

hv_netvsc: Implement batching in send buffer

With this patch, we can send out multiple RNDIS data packets in one send buffer
slot and one VMBus message. It reduces the overhead associated with VMBus messages.

Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

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 your net-next tree.
Basically, nf_tables updates to add the set extension infrastructure and finish
the transaction for sets from Patrick McHardy. More specifically, they are:

1) Move netns to basechain and use recently added possible_net_t, from
   Patrick McHardy.

2) Use LOGLEVEL_<FOO> from nf_log infrastructure, from Joe Perches.

3) Restore nf_log_trace that was accidentally removed during conflict
   resolution.

4) nft_queue does not depend on NETFILTER_XTABLES, starting from here
   all patches from Patrick McHardy.

5) Use raw_smp_processor_id() in nft_meta.

Then, several patches to prepare ground for the new set extension
infrastructure:

6) Pass object length to the hash callback in rhashtable as needed by
   the new set extension infrastructure.

7) Cleanup patch to restore struct nft_hash as wrapper for struct
   rhashtable

8) Another small source code readability cleanup for nft_hash.

9) Convert nft_hash to rhashtable callbacks.

And finally...

10) Add the new set extension infrastructure.

11) Convert the nft_hash and nft_rbtree sets to use it.

12) Batch set element release to avoid several RCU grace period in a row
    and add new function nft_set_elem_destroy() to consolidate set element
    release.

13) Return the set extension data area from nft_lookup.

14) Refactor existing transaction code to add some helper functions
    and document it.

15) Complete the set transaction support, using similar approach to what we
    already use, to activate/deactivate elements in an atomic fashion.
====================

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

Merge branch 'tipc-next'

Ying Xue says:

====================
tipc: fix two corner issues

The patch set aims at resolving the following two critical issues:

Patch #1: Resolve a deadlock which happens while all links are reset
Patch #2: Correct a mistake usage of RCU lock which is used to protect
          node list
====================

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

tipc: involve reference counter for node structure

TIPC node hash node table is protected with rcu lock on read side.
tipc_node_find() is used to look for a node object with node address
through iterating the hash node table. As the entire process of what
tipc_node_find() traverses the table is guarded with rcu read lock,
it's safe for us. However, when callers use the node object returned
by tipc_node_find(), there is no rcu read lock applied. Therefore,
this is absolutely unsafe for callers of tipc_node_find().

Now we introduce a reference counter for node structure. Before
tipc_node_find() returns node object to its caller, it first increases
the reference counter. Accordingly, after its caller used it up,
it decreases the counter again. This can prevent a node being used by
one thread from being freed by another thread.

Reviewed-by: Erik Hugne <erik.hugne@ericsson.com>
Reviewed-by: Jon Maloy <jon.maloy@ericson.com>
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tipc: fix potential deadlock when all links are reset

[   60.988363] ======================================================
[   60.988754] [ INFO: possible circular locking dependency detected ]
[   60.989152] 3.19.0+ #194 Not tainted
[   60.989377] -------------------------------------------------------
[   60.989781] swapper/3/0 is trying to acquire lock:
[   60.990079]  (&(&n_ptr->lock)->rlock){+.-...}, at: [<ffffffffa0006dca>] tipc_link_retransmit+0x1aa/0x240 [tipc]
[   60.990743]
[   60.990743] but task is already holding lock:
[   60.991106]  (&(&bclink->lock)->rlock){+.-...}, at: [<ffffffffa00004be>] tipc_bclink_lock+0x8e/0xa0 [tipc]
[   60.991738]
[   60.991738] which lock already depends on the new lock.
[   60.991738]
[   60.992174]
[   60.992174] the existing dependency chain (in reverse order) is:
[   60.992174]
-> #1 (&(&bclink->lock)->rlock){+.-...}:
[   60.992174]        [<ffffffff810a9c0c>] lock_acquire+0x9c/0x140
[   60.992174]        [<ffffffff8179c41f>] _raw_spin_lock_bh+0x3f/0x50
[   60.992174]        [<ffffffffa00004be>] tipc_bclink_lock+0x8e/0xa0 [tipc]
[   60.992174]        [<ffffffffa0000f57>] tipc_bclink_add_node+0x97/0xf0 [tipc]
[   60.992174]        [<ffffffffa0011815>] tipc_node_link_up+0xf5/0x110 [tipc]
[   60.992174]        [<ffffffffa0007783>] link_state_event+0x2b3/0x4f0 [tipc]
[   60.992174]        [<ffffffffa00193c0>] tipc_link_proto_rcv+0x24c/0x418 [tipc]
[   60.992174]        [<ffffffffa0008857>] tipc_rcv+0x827/0xac0 [tipc]
[   60.992174]        [<ffffffffa0002ca3>] tipc_l2_rcv_msg+0x73/0xd0 [tipc]
[   60.992174]        [<ffffffff81646e66>] __netif_receive_skb_core+0x746/0x980
[   60.992174]        [<ffffffff816470c1>] __netif_receive_skb+0x21/0x70
[   60.992174]        [<ffffffff81647295>] netif_receive_skb_internal+0x35/0x130
[   60.992174]        [<ffffffff81648218>] napi_gro_receive+0x158/0x1d0
[   60.992174]        [<ffffffff81559e05>] e1000_clean_rx_irq+0x155/0x490
[   60.992174]        [<ffffffff8155c1b7>] e1000_clean+0x267/0x990
[   60.992174]        [<ffffffff81647b60>] net_rx_action+0x150/0x360
[   60.992174]        [<ffffffff8105ec43>] __do_softirq+0x123/0x360
[   60.992174]        [<ffffffff8105f12e>] irq_exit+0x8e/0xb0
[   60.992174]        [<ffffffff8179f9f5>] do_IRQ+0x65/0x110
[   60.992174]        [<ffffffff8179da6f>] ret_from_intr+0x0/0x13
[   60.992174]        [<ffffffff8100de9f>] arch_cpu_idle+0xf/0x20
[   60.992174]        [<ffffffff8109dfa6>] cpu_startup_entry+0x2f6/0x3f0
[   60.992174]        [<ffffffff81033cda>] start_secondary+0x13a/0x150
[   60.992174]
-> #0 (&(&n_ptr->lock)->rlock){+.-...}:
[   60.992174]        [<ffffffff810a8f7d>] __lock_acquire+0x163d/0x1ca0
[   60.992174]        [<ffffffff810a9c0c>] lock_acquire+0x9c/0x140
[   60.992174]        [<ffffffff8179c41f>] _raw_spin_lock_bh+0x3f/0x50
[   60.992174]        [<ffffffffa0006dca>] tipc_link_retransmit+0x1aa/0x240 [tipc]
[   60.992174]        [<ffffffffa0001e11>] tipc_bclink_rcv+0x611/0x640 [tipc]
[   60.992174]        [<ffffffffa0008646>] tipc_rcv+0x616/0xac0 [tipc]
[   60.992174]        [<ffffffffa0002ca3>] tipc_l2_rcv_msg+0x73/0xd0 [tipc]
[   60.992174]        [<ffffffff81646e66>] __netif_receive_skb_core+0x746/0x980
[   60.992174]        [<ffffffff816470c1>] __netif_receive_skb+0x21/0x70
[   60.992174]        [<ffffffff81647295>] netif_receive_skb_internal+0x35/0x130
[   60.992174]        [<ffffffff81648218>] napi_gro_receive+0x158/0x1d0
[   60.992174]        [<ffffffff81559e05>] e1000_clean_rx_irq+0x155/0x490
[   60.992174]        [<ffffffff8155c1b7>] e1000_clean+0x267/0x990
[   60.992174]        [<ffffffff81647b60>] net_rx_action+0x150/0x360
[   60.992174]        [<ffffffff8105ec43>] __do_softirq+0x123/0x360
[   60.992174]        [<ffffffff8105f12e>] irq_exit+0x8e/0xb0
[   60.992174]        [<ffffffff8179f9f5>] do_IRQ+0x65/0x110
[   60.992174]        [<ffffffff8179da6f>] ret_from_intr+0x0/0x13
[   60.992174]        [<ffffffff8100de9f>] arch_cpu_idle+0xf/0x20
[   60.992174]        [<ffffffff8109dfa6>] cpu_startup_entry+0x2f6/0x3f0
[   60.992174]        [<ffffffff81033cda>] start_secondary+0x13a/0x150
[   60.992174]
[   60.992174] other info that might help us debug this:
[   60.992174]
[   60.992174]  Possible unsafe locking scenario:
[   60.992174]
[   60.992174]        CPU0                    CPU1
[   60.992174]        ----                    ----
[   60.992174]   lock(&(&bclink->lock)->rlock);
[   60.992174]                                lock(&(&n_ptr->lock)->rlock);
[   60.992174]                                lock(&(&bclink->lock)->rlock);
[   60.992174]   lock(&(&n_ptr->lock)->rlock);
[   60.992174]
[   60.992174]  *** DEADLOCK ***
[   60.992174]
[   60.992174] 3 locks held by swapper/3/0:
[   60.992174]  #0:  (rcu_read_lock){......}, at: [<ffffffff81646791>] __netif_receive_skb_core+0x71/0x980
[   60.992174]  #1:  (rcu_read_lock){......}, at: [<ffffffffa0002c35>] tipc_l2_rcv_msg+0x5/0xd0 [tipc]
[   60.992174]  #2:  (&(&bclink->lock)->rlock){+.-...}, at: [<ffffffffa00004be>] tipc_bclink_lock+0x8e/0xa0 [tipc]
[   60.992174]

The correct the sequence of grabbing n_ptr->lock and bclink->lock
should be that the former is first held and the latter is then taken,
which exactly happened on CPU1. But especially when the retransmission
of broadcast link is failed, bclink->lock is first held in
tipc_bclink_rcv(), and n_ptr->lock is taken in link_retransmit_failure()
called by tipc_link_retransmit() subsequently, which is demonstrated on
CPU0. As a result, deadlock occurs.

If the order of holding the two locks happening on CPU0 is reversed, the
deadlock risk will be relieved. Therefore, the node lock taken in
link_retransmit_failure() originally is moved to tipc_bclink_rcv()
so that it's obtained before bclink lock. But the precondition of
the adjustment of node lock is that responding to bclink reset event
must be moved from tipc_bclink_unlock() to tipc_node_unlock().

Reviewed-by: Erik Hugne <erik.hugne@ericsson.com>
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

virtio: simplify the using of received in virtnet_poll

received is 0, no need to minus it and use "+=" to reassign it

Signed-off-by: Li RongQing <roy.qing.li@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'be2net-next'

Sathya Perla says:

====================
be2net: patch set

Hi David, this patch set includes 2 feature additions to the be2net driver:

Patch 1 sets up cpu affinity hints for be2net irqs using the
cpumask_set_cpu_local_first() API that first picks the near numa cores
and when they are exhausted, selects the far numa cores.

Patch 2 setups up xps queue mapping for be2net's TXQs to avoid,
by default, TX lock contention.

Patch 3 just bumps up the driver version.

Pls consider applying this patch set to the net-next queue. Thanks!
====================

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

be2net: bump up the driver version to 10.6.0.1

Signed-off-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: setup xps queue mapping

This patch sets up xps queue mapping on load, so that TX traffic is
steered to the queue whose irqs are being processed by the current cpu.
This helps in avoiding TX lock contention.

Signed-off-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: assign CPU affinity hints to be2net IRQs

This patch provides hints to irqbalance to map be2net IRQs to
specific CPU cores. cpumask_set_cpu_local_first() is used, which first
maps IRQs to near NUMA cores; when those cores are exhausted, IRQs are
mapped to far NUMA cores.

Signed-off-by: Padmanabh Ratnakar <padmanabh.ratnakar@emulex.com>
Signed-off-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: tcp_syn_flood_action() can be static

After commit 1fb6f159fd21 ("tcp: add tcp_conn_request"),
tcp_syn_flood_action() is no longer used from IPv6.

We can make it static, by moving it above tcp_conn_request()

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Octavian Purdila <octavian.purdila@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb4: fix boolreturn.cocci warnings

drivers/net/ethernet/chelsio/cxgb4/cxgb4_fcoe.c:49:9-10: WARNING: return of 0/1 in function 'cxgb_fcoe_sof_eof_supported' with return type bool

 Return statements in functions returning bool should use
 true/false instead of 1/0.
Generated by: scripts/coccinelle/misc/boolreturn.cocci

CC: Varun Prakash <varun@chelsio.com>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

fib6: install fib6 ops in the last step

We should not commit the new ops until we finish
all the setup, otherwise we have to NULL it on failure.

Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'bcmgenet-next'

Petri Gynther says:

====================
net: bcmgenet: multiple Rx queues support

Final patch set to add support for multiple Rx queues:
1. remove priv->int0_mask and priv->int1_mask
2. modify Tx ring int_enable and int_disable vectors
3. simplify bcmgenet_init_dma()
4. tweak init_umac()
5. rework Tx NAPI code
6. rework Rx NAPI code
7. add support for multiple Rx queues
====================

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

net: bcmgenet: add support for multiple Rx queues

Add support for multiple Rx queues:
1. Add NAPI context per Rx queue
2. Modify Rx interrupt and Rx NAPI code to handle multiple Rx queues

Signed-off-by: Petri Gynther <pgynther@google.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: bcmgenet: rework Rx NAPI code

Introduce new bcmgenet functions to handle the NAPI calls to:
netif_napi_add()
napi_enable()
napi_disable()
netif_napi_del()

Signed-off-by: Petri Gynther <pgynther@google.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: bcmgenet: rework Tx NAPI code

Introduce new bcmgenet functions to handle the NAPI calls to:
netif_napi_add()
napi_enable()
napi_disable()
netif_napi_del()

Signed-off-by: Petri Gynther <pgynther@google.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: bcmgenet: tweak init_umac()

Use more meaningful variable names int0_enable and int1_enable when
enabling bcmgenet interrupts.

For Rx default queue interrupts, use:
UMAC_IRQ_RXDMA_BDONE | UMAC_IRQ_RXDMA_PDONE

For Tx default queue interrupts, use:
UMAC_IRQ_TXDMA_BDONE | UMAC_IRQ_TXDMA_PDONE

Signed-off-by: Petri Gynther <pgynther@google.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: bcmgenet: simplify bcmgenet_init_dma()

Do the two kcalloc() calls first, before proceeding into Rx/Tx DMA init.
Makes the error case handling much simpler.

Signed-off-by: Petri Gynther <pgynther@google.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Jaedon Shin <jaedon.shin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: bcmgenet: modify Tx ring int_enable and int_disable vectors

Remove unnecessary function parameter priv. Use ring->priv instead.

Signed-off-by: Petri Gynther <pgynther@google.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: bcmgenet: remove priv->int0_mask and priv->int1_mask

Remove unused priv->int0_mask and priv->int1_mask.

Signed-off-by: Petri Gynther <pgynther@google.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'xgene-next'

Iyappan Subramanian says:

====================
drivers: net: xgene: Add separate tx completion ring

SGMII based 1GbE and 10GbE interfaces support multiple interrupts.
Adding separate tx completion descriptor ring and associating a dedicated irq for the TX completion.
====================

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: Keyur Chudgar <kchudgar@apm.com>

drivers: net: xgene: Add separate tx completion ring

- Added wrapper functions around napi_add, napi_del, napi_enable and napi_disable
- Moved platform_get_irq function call after reading phy_mode
- Associating the new irq to tx completion for the supported ethernet interfaces

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: Keyur Chudgar <kchudgar@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

dtb: xgene: Add interrupt for Tx completion

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: Keyur Chudgar <kchudgar@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Documentation: dts: xgene: Update interrupt field description

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: Keyur Chudgar <kchudgar@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netfilter: nf_tables: implement set transaction support

Set elements are the last object type not supporting transaction support.
Implement similar to the existing rule transactions:

The global transaction counter keeps track of two generations, current
and next. Each element contains a bitmask specifying in which generations
it is inactive.

New elements start out as inactive in the current generation and active
in the next. On commit, the previous next generation becomes the current
generation and the element becomes active. The bitmask is then cleared
to indicate that the element is active in all future generations. If the
transaction is aborted, the element is removed from the set before it
becomes active.

When removing an element, it gets marked as inactive in the next generation.
On commit the next generation becomes active and the therefor the element
inactive. It is then taken out of then set and released. On abort, the
element is marked as active for the next generation again.

Lookups ignore elements not active in the current generation.

The current set types (hash/rbtree) both use a field in the extension area
to store the generation mask. This (currently) does not require any
additional memory since we have some free space in there.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

netfilter: nf_tables: add transaction helper functions

Add some helper functions for building the genmask as preparation for
set transactions.

Also add a little documentation how this stuff actually works.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

netfilter: nf_tables: return set extensions from ->lookup()

Return the extension area from the ->lookup() function to allow to
consolidate common actions.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

netfilter: nf_tables: consolide set element destruction

With the conversion to set extensions, it is now possible to consolidate
the different set element destruction functions.

The set implementations' ->remove() functions are changed to only take
the element out of their internal data structures. Elements will be freed
in a batched fashion after the global transaction's completion RCU grace
period.

This reduces the amount of grace periods required for nft_hash from N
to zero additional ones, additionally this guarantees that the set
elements' extensions of all implementations can be used under RCU
protection.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

ipv6: hash net ptr into fragmentation bucket selection

As namespaces are sometimes used with overlapping ip address ranges,
we should also use the namespace as input to the hash to select the ip
fragmentation counter bucket.

Cc: Eric Dumazet <edumazet@google.com>
Cc: Flavio Leitner <fbl@redhat.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: hash net ptr into fragmentation bucket selection

As namespaces are sometimes used with overlapping ip address ranges,
we should also use the namespace as input to the hash to select the ip
fragmentation counter bucket.

Cc: Eric Dumazet <edumazet@google.com>
Cc: Flavio Leitner <fbl@redhat.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'tipc-next'

Jon Maloy says:

====================
tipc: some improvements and fixes

We introduce a better algorithm for selecting when and which
users should be subject to link congestion control, plus clean
up some code for that mechanism.
Commit #3 fixes another rare race condition during packet reception.
====================

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

tipc: eliminate race condition at dual link establishment

Despite recent improvements, the establishment of dual parallel
links still has a small glitch where messages can bypass each
other. When the second link in a dual-link configuration is
established, part of the first link's traffic will be steered over
to the new link. Although we do have a mechanism to ensure that
packets sent before and after the establishment of the new link
arrive in sequence to the destination node, this is not enough.
The arriving messages will still be delivered upwards in different
threads, something entailing a risk of message disordering during
the transition phase.

To fix this, we introduce a synchronization mechanism between the
two parallel links, so that traffic arriving on the new link cannot
be added to its input queue until we are guaranteed that all
pre-establishment messages have been delivered on the old, parallel
link.

This problem seems to always have been around, but its occurrence is
so rare that it has not been noticed until recent intensive testing.

Reviewed-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Erik Hugne <erik.hugne@ericsson.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tipc: clean up handling of link congestion

After the recent changes in message importance handling it becomes
possible to simplify handling of messages and sockets when we
encounter link congestion.

We merge the function tipc_link_cong() into link_schedule_user(),
and simplify the code of the latter. The code should now be
easier to follow, especially regarding return codes and handling
of the message that caused the situation.

In case the scheduling function is unable to pre-allocate a wakeup
message buffer, it now returns -ENOBUFS, which is a more correct
code than the previously used -EHOSTUNREACH.

Reviewed-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Erik Hugne <erik.hugne@ericsson.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tipc: introduce starvation free send algorithm

Currently, we only use a single counter; the length of the backlog
queue, to determine whether a message should be accepted to the queue
or not. Each time a message is being sent, the queue length is compared
to a threshold value for the message's importance priority. If the queue
length is beyond this threshold, the message is rejected. This algorithm
implies a risk of starvation of low importance senders during very high
load, because it may take a long time before the backlog queue has
decreased enough to accept a lower level message.

We now eliminate this risk by introducing a counter for each importance
priority. When a message is sent, we check only the queue level for that
particular message's priority. If that is ok, the message can be added
to the backlog, irrespective of the queue level for other priorities.
This way, each level is guaranteed a certain portion of the total
bandwidth, and any risk of starvation is eliminated.

Reviewed-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Erik Hugne <erik.hugne@ericsson.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: Handle non-bridge master change

Master change notifications may occur other than when joining or
leaving a bridge, for example when being added to or removed from
a bond or Open vSwitch. In that case, do nothing instead of asking
the switch driver to remove a port from a bridge that it didn't join.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netfilter: nf_tables: convert hash and rbtree to set extensions

The set implementations' private struct will only contain the elements
needed to maintain the search structure, all other elements are moved
to the set extensions.

Element allocation and initialization is performed centrally by
nf_tables_api instead of by the different set implementations'
->insert() functions. A new "elemsize" member in the set ops specifies
the amount of memory to reserve for internal usage. Destruction
will also be moved out of the set implementations by a following patch.

Except for element allocation, the patch is a simple conversion to
using data from the extension area.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

netfilter: nf_tables: add set extensions

Add simple set extension infrastructure for maintaining variable sized
and optional per element data.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

netfilter: nft_hash: convert to use rhashtable callbacks

A following patch will convert sets to use so called set extensions,
where the key is not located in a fixed position anymore. This will
require rhashtable hashing and comparison callbacks to be used.

As preparation, convert nft_hash to use these callbacks without any
functional changes.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

netfilter: nft_hash: indent rhashtable parameters

Improve readability by indenting the parameter initialization.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

netfilter: nft_hash: restore struct nft_hash

Following patches will add new private members, restore struct nft_hash
as preparation.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

rhashtable: provide len to obj_hashfn

nftables sets will be converted to use so called setextensions, moving
the key to a non-fixed position. To hash it, the obj_hashfn must be used,
however it so far doesn't receive the length parameter.

Pass the key length to obj_hashfn() and convert existing users.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

crypto: algif - fix warn: unsigned 'used' is never less than zero

Change type from unsigned long to int to fix an issue reported by kbuild robot:
crypto/algif_skcipher.c:596 skcipher_recvmsg_async() warn: unsigned 'used' is
never less than zero.

Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tipc: fix a link reset issue due to retransmission failures

When a node joins a cluster while we are transmitting a fragment
stream over the broadcast link, it's missing the preceding fragments
needed to build a meaningful message. As a result, the node has to
drop it. However, as the fragment message is not acknowledged to
its sender before it's dropped, it accidentally causes link reset
of retransmission failure on the node.

Reported-by: Erik Hugne <erik.hugne@ericsson.com>
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Erik Hugne <erik.hugne@ericsson.com>
Tested-by: Erik Hugne <erik.hugne@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

s390: fix /proc/interrupts output

The irqclass_sub_desc array and enum interruption_class are out of sync
thus /proc/interrupts is broken. Remove IRQIO_CLW.

Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Ursula Braun <ursula.braun@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sctp: avoid to repeatedly declare external variables

Move the declaration for external variables to sctp.h file avoiding
to repeatedly declare them with extern keyword.

Signed-off-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netfilter: nft_meta: use raw_smp_processor_id()

Using smp_processor_id() triggers warnings with PREEMPT_RCU. There is no
point in disabling preemption since we only collect the numeric value,
so use raw_smp_processor_id() instead.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

netfilter: nf_tables: nft_queue does not depend on x_tables

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

netfilter: nf_tables: restore nf_log_trace() in nf_tables_core.c

As described by 4017a7e ("netfilter: restore rule tracing via
nfnetlink_log"), this accidentally slipped through during conflict
resolution in d5c1d8c.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

netfilter: Use LOGLEVEL_<FOO> defines

Use the #defines where appropriate.

Miscellanea:

Add explicit #include <linux/kernel.h> where it was not
previously used so that these #defines are a bit more
explicitly defined instead of indirectly included via:
module.h->moduleparam.h->kernel.h

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

netfilter: nf_tables: move struct net pointer to base chain

The network namespace is only needed for base chains to get at the
gencursor. Also convert to possible_net_t.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

tcp: fix ipv4 mapped request socks

ss should display ipv4 mapped request sockets like this :

tcp    SYN-RECV   0      0  ::ffff:192.168.0.1:8080   ::ffff:192.0.2.1:35261

and not like this :

tcp    SYN-RECV   0      0  192.168.0.1:8080   192.0.2.1:35261

We should init ireq->ireq_family based on listener sk_family,
not the actual protocol carried by SYN packet.

This means we can set ireq_family in inet_reqsk_alloc()

Fixes: 3f66b083a5b7 ("inet: introduce ireq_family")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

virtio: change comment in transmit

The original comment was not really informative or funny
as well as sexist. Replace it with a better explanation of
why the driver does stop and what the impacts are.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

tools: bpf_asm: cleanup vlan extension related token

We now have K_VLANT, K_VLANP and K_VLANTPID. Clean them up into more
descriptive token, namely K_VLAN_TCI, K_VLAN_AVAIL and K_VLAN_TPID.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'listener_refactor_16'

Eric Dumazet says:

====================
tcp: listener refactor part 16

A CONFIG_PROVE_RCU=y build revealed an RCU splat I had to fix.

I added const qualifiers to various md5 methods, as I expect
to call them on behalf of request sock traffic even if
the listener socket is not locked. This seems ok, but adding
const makes the contract clearer. Note a good reduction
of code size thanks to request/establish sockets convergence.
====================

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

tcp: md5: get rid of tcp_v[46]_reqsk_md5_lookup()

With request socks convergence, we no longer need
different lookup methods. A request socket can
use generic lookup function.

Add const qualifier to 2nd tcp_v[46]_md5_lookup() parameter.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: md5: remove request sock argument of calc_md5_hash()

Since request and established sockets now have same base,
there is no need to pass two pointers to tcp_v4_md5_hash_skb()
or tcp_v6_md5_hash_skb()

Also add a const qualifier to their struct tcp_md5sig_key argument.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: md5: input path is run under rcu protected sections

It is guaranteed that both tcp_v4_rcv() and tcp_v6_rcv()
run from rcu read locked sections :

ip_local_deliver_finish() and ip6_input_finish() both
use rcu_read_lock()

Also align tcp_v6_inbound_md5_hash() on tcp_v4_inbound_md5_hash()
by returning a boolean.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: use C99 initializers in new_state[]

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: md5: fix rcu lockdep splat

While timer handler effectively runs a rcu read locked section,
there is no explicit rcu_read_lock()/rcu_read_unlock() annotations
and lockdep can be confused here :

net/ipv4/tcp_ipv4.c-906-        /* caller either holds rcu_read_lock() or socket lock */
net/ipv4/tcp_ipv4.c:907:        md5sig = rcu_dereference_check(tp->md5sig_info,
net/ipv4/tcp_ipv4.c-908-                                       sock_owned_by_user(sk) ||
net/ipv4/tcp_ipv4.c-909-                                       lockdep_is_held(&sk->sk_lock.slock));

Let's explicitely acquire rcu_read_lock() in tcp_make_synack()

Before commit fa76ce7328b ("inet: get rid of central tcp/dccp listener
timer"), we were holding listener lock so lockdep was happy.

Fixes: fa76ce7328b ("inet: get rid of central tcp/dccp listener timer")
Signed-off-by: Eric DUmazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'rhashtable-next'

Thomas Graf says:

====================
rhashtable updates on top of Herbert's work

Patch 1 is a bugfix for an RCU splash I encountered while testing.
Patch 2 & 3 are pure cleanups. Patch 4 disables automatic shrinking
by default as discussed in previous thread. Patch 5 removes some
rhashtable internal knowledge from nft_hash and fixes another RCU
splash.

I've pushed various rhashtable tests (Netlink, nft) together with a
Makefile to a git tree [0] for easier stress testing.

[0] https://github.com/tgraf/rhashtable
====================

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

rhashtable: Add rhashtable_free_and_destroy()

rhashtable_destroy() variant which stops rehashes, iterates over
the table and calls a callback to release resources.

Avoids need for nft_hash to embed rhashtable internals and allows to
get rid of the being_destroyed flag. It also saves a 2nd mutex
lock upon destruction.

Also fixes an RCU lockdep splash on nft set destruction due to
calling rht_for_each_entry_safe() without holding bucket locks.
Open code this loop as we need know that no mutations may occur in
parallel.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

rhashtable: Disable automatic shrinking by default

Introduce a new bool automatic_shrinking to require the
user to explicitly opt-in to automatic shrinking of tables.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

rhashtable: Mark internal/private inline functions as such

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

rhashtable: Use 'unsigned int' consistently

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

rhashtable: Extend RCU read lock into rhashtable_insert_rehash()

rhashtable_insert_rehash() requires RCU locks to be held in order
to access ht->tbl and traverse to the last table.

Fixes: ccd57b1bd324 ("rhashtable: Add immediate rehash during insertion")
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

filter: introduce SKF_AD_VLAN_TPID BPF extension

If vlan offloading takes place then vlan header is removed from frame
and its contents, both vlan_tci and vlan_proto, is available to user
space via TPACKET interface. However, only vlan_tci can be used in BPF
filters.

This commit introduces a new BPF extension. It makes possible to load
the value of vlan_proto (vlan TPID) to register A. Support for classic
BPF and eBPF is being added, analogous to skb->protocol.

Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: Michal Sekletar <msekleta@redhat.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Reviewed-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'cxgb4-fcoe'

Varun Prakash says:

====================
FCoE support in cxgb4 driver

This patch series enables FCoE support in cxgb4 driver, it enables
FCOE_CRC and FCOE_MTU net device features.

This series is created against net-next tree.
====================

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

cxgb4: update Kconfig and Makefile for FCoE support

Signed-off-by: Varun Prakash <varun@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb4: add cxgb4_fcoe.c for FCoE

This patch adds cxgb4_fcoe.c and enables FCOE_CRC, FCOE_MTU
net device features.

Signed-off-by: Varun Prakash <varun@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb4: add cxgb4_fcoe.h and macro definitions for FCoE

This patch adds new header file cxgb4_fcoe.h and defines new
macros for FCoE support in cxgb4 driver.

Signed-off-by: Varun Prakash <varun@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: fix sparse warnings in privacy stable addresses generation

Those warnings reported by sparse endianness check (via kbuild test robot)
are harmless, nevertheless fix them up and make the code a little bit
easier to read.

Reported-by: kbuild test robot <fengguang.wu@intel.com>
Fixes: 622c81d57b392cc ("ipv6: generation of stable privacy addresses for link-local and autoconf")
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

tipc: fix compile error when IPV6=m and TIPC=y

When IPV6=m and TIPC=y, below error will appear during building kernel
image:

net/tipc/udp_media.c:196:
undefined reference to `ip6_dst_lookup'
make: *** [vmlinux] Error 1

As ip6_dst_lookup() is implemented in IPV6 and IPV6 is compiled as
module, ip6_dst_lookup() is not built-in core kernel image. As a
result, compiler cannot find 'ip6_dst_lookup' reference while
compiling TIPC code into core kernel image.

But with the method introduced by commit 5f81bd2e5d80 ("ipv6: export a
stub for IPv6 symbols used by vxlan"), we can avoid the compile error
through "ipv6_stub" pointer to access ip6_dst_lookup().

Fixes: d0f91938bede ("tipc: add ip/udp media type")
Suggested-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: allow to delete a whole device group

With dev group, we can change a batch of net devices,
so we should allow to delete them together too.

Group 0 is not allowed to be deleted since it is
the default group.

Cc: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

rhashtable: Add comment on choice of elasticity value

This patch adds a comment on the choice of the value 16 as the
maximum chain length before we force a rehash.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

cx82310_eth: fix semicolon.cocci warnings

drivers/net/usb/cx82310_eth.c:175:2-3: Unneeded semicolon

 Removes unneeded semicolon.

Generated by: scripts/coccinelle/misc/semicolon.cocci

CC: Ondrej Zary <linux@rainbow-software.org>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tipc: validate length of sockaddr in connect() for dgram/rdm

Commit f2f8036 ("tipc: add support for connect() on dgram/rdm sockets")
hasn't validated user input length for the sockaddr structure which allows
a user to overwrite kernel memory with arbitrary input.

Fixes: f2f8036 ("tipc: add support for connect() on dgram/rdm sockets")
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: remove never used forwarding_accel_ops pointer from net_device

Cc: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge git://git./linux/kernel/git/davem/net

Conflicts:
net/netfilter/nf_tables_core.c

The nf_tables_core.c conflict was resolved using a conflict resolution
from Stephen Rothwell as a guide.

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

rhashtable: Fix sleeping inside RCU critical section in walk_stop

The commit 963ecbd41a1026d99ec7537c050867428c397b89 ("rhashtable:
Fix use-after-free in rhashtable_walk_stop") fixed a real bug
but created another one because we may end up sleeping inside an
RCU critical section.

This patch fixes it properly by replacing the mutex with a spin
lock that specifically protects the walker lists.

Reported-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'ipv6_stable_privacy_address'

Hannes Frederic Sowa says:

====================
ipv6: RFC7217 stable privacy addresses implementation

this is an implementation of basic support for RFC7217 stable privacy
addresses. Please review and consider for net-next.

v2:
* Correct references to RFC 7212 -> RFC 7217 in documentation patch (thanks, Eric!)
====================

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

ipv6: add documentation for stable_secret, idgen_delay and idgen_retries knobs

Cc: Erik Kline <ek@google.com>
Cc: Fernando Gont <fgont@si6networks.com>
Cc: Lorenzo Colitti <lorenzo@google.com>
Cc: YOSHIFUJI Hideaki/吉藤英明 <hideaki.yoshifuji@miraclelinux.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: introduce idgen_delay and idgen_retries knobs

This is specified by RFC 7217.

Cc: Erik Kline <ek@google.com>
Cc: Fernando Gont <fgont@si6networks.com>
Cc: Lorenzo Colitti <lorenzo@google.com>
Cc: YOSHIFUJI Hideaki/吉藤英明 <hideaki.yoshifuji@miraclelinux.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: do retries on stable privacy addresses

If a DAD conflict is detected, we want to retry privacy stable address
generation up to idgen_retries (= 3) times with a delay of idgen_delay
(= 1 second). Add the logic to addrconf_dad_failure.

By design, we don't clean up dad failed permanent addresses.

Cc: Erik Kline <ek@google.com>
Cc: Fernando Gont <fgont@si6networks.com>
Cc: Lorenzo Colitti <lorenzo@google.com>
Cc: YOSHIFUJI Hideaki/吉藤英明 <hideaki.yoshifuji@miraclelinux.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: collapse state_lock and lock

Cc: Erik Kline <ek@google.com>
Cc: Fernando Gont <fgont@si6networks.com>
Cc: Lorenzo Colitti <lorenzo@google.com>
Cc: YOSHIFUJI Hideaki/吉藤英明 <hideaki.yoshifuji@miraclelinux.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: introduce IFA_F_STABLE_PRIVACY flag

We need to mark appropriate addresses so we can do retries in case their
DAD failed.

Cc: Erik Kline <ek@google.com>
Cc: Fernando Gont <fgont@si6networks.com>
Cc: Lorenzo Colitti <lorenzo@google.com>
Cc: YOSHIFUJI Hideaki/吉藤英明 <hideaki.yoshifuji@miraclelinux.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>