Skip to content

L
linux
Switch branch/tag
  1. 09 Nov, 2018 7 commits
  3. 05 Nov, 2018 2 commits
    • Jarod Wilson's avatar
      bonding/802.3ad: fix link_failure_count tracking · ea53abfa
      Jarod Wilson authored 
      Commit 4d2c0cda set slave->link to
BOND_LINK_DOWN for 802.3ad bonds whenever invalid speed/duplex values
were read, to fix a problem with slaves getting into weird states, but
in the process, broke tracking of link failures, as going straight to
BOND_LINK_DOWN when a link is indeed down (cable pulled, switch rebooted)
means we broke out of bond_miimon_inspect()'s BOND_LINK_DOWN case because
!link_state was already true, we never incremented commit, and never got
a chance to call bond_miimon_commit(), where slave->link_failure_count
would be incremented. I believe the simple fix here is to mark the slave
as BOND_LINK_FAIL, and let bond_miimon_inspect() transition the link from
_FAIL to either _UP or _DOWN, and in the latter case, we now get proper
incrementing of link_failure_count again.

Fixes: 4d2c0cda ("bonding: speed/duplex update at NETDEV_UP event")
CC: Mahesh Bandewar <maheshb@google.com>
CC: David S. Miller <davem@davemloft.net>
CC: netdev@vger.kernel.org
CC: stable@vger.kernel.org
Signed-off-by: default avatarJarod Wilson <jarod@redhat.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      ea53abfa
    • Holger Hoffstätte's avatar
      net: phy: realtek: fix RTL8201F sysfs name · 0432e833
      Holger Hoffstätte authored 
      Since 4.19 the following error in sysfs has appeared when using the
r8169 NIC driver:

$cd /sys/module/realtek/drivers
$ls -l
ls: cannot access 'mdio_bus:RTL8201F 10/100Mbps Ethernet': No such file or directory
[..garbled dir entries follow..]

Apparently the forward slash in "10/100Mbps Ethernet" is interpreted
as directory separator that leads nowhere, and was introduced in commit
513588dd ("net: phy: realtek: add RTL8201F phy-id and functions").

Fix this by removing the offending slash in the driver name.

Other drivers in net/phy seem to have the same problem, but I cannot
test/verify them.

Fixes: 513588dd ("net: phy: realtek: add RTL8201F phy-id and functions")
Signed-off-by: default avatarHolger Hoffstätte <holger@applied-asynchrony.com>
Reviewed-by: default avatarAndrew Lunn <andrew@lunn.ch>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      0432e833
  5. 04 Nov, 2018 6 commits
  7. 03 Nov, 2018 20 commits
    • Yunsheng Lin's avatar
      net: hns3: Fix for out-of-bounds access when setting pfc back pressure · e8ccbb7d
      Yunsheng Lin authored 
      The vport should be initialized to hdev->vport for each bp group,
otherwise it will cause out-of-bounds access and bp setting not
correct problem.

[   35.254124] BUG: KASAN: slab-out-of-bounds in hclge_pause_setup_hw+0x2a0/0x3f8 [hclge]
[   35.254126] Read of size 2 at addr ffff803b6651581a by task kworker/0:1/14

[   35.254132] CPU: 0 PID: 14 Comm: kworker/0:1 Not tainted 4.19.0-rc7-hulk+ #85
[   35.254133] Hardware name: Huawei D06/D06, BIOS Hisilicon D06 UEFI RC0 - B052 (V0.52) 09/14/2018
[   35.254141] Workqueue: events work_for_cpu_fn
[   35.254144] Call trace:
[   35.254147]  dump_backtrace+0x0/0x2f0
[   35.254149]  show_stack+0x24/0x30
[   35.254154]  dump_stack+0x110/0x184
[   35.254157]  print_address_description+0x168/0x2b0
[   35.254160]  kasan_report+0x184/0x310
[   35.254162]  __asan_load2+0x7c/0xa0
[   35.254170]  hclge_pause_setup_hw+0x2a0/0x3f8 [hclge]
[   35.254177]  hclge_tm_init_hw+0x794/0x9f0 [hclge]
[   35.254184]  hclge_tm_schd_init+0x48/0x58 [hclge]
[   35.254191]  hclge_init_ae_dev+0x778/0x1168 [hclge]
[   35.254196]  hnae3_register_ae_dev+0x14c/0x298 [hnae3]
[   35.254206]  hns3_probe+0x88/0xa8 [hns3]
[   35.254210]  local_pci_probe+0x7c/0xf0
[   35.254212]  work_for_cpu_fn+0x34/0x50
[   35.254214]  process_one_work+0x4d4/0xa38
[   35.254216]  worker_thread+0x55c/0x8d8
[   35.254219]  kthread+0x1b0/0x1b8
[   35.254222]  ret_from_fork+0x10/0x1c

[   35.254224] The buggy address belongs to the page:
[   35.254228] page:ffff7e00ed994400 count:1 mapcount:0 mapping:0000000000000000 index:0x0 compound_mapcount: 0
[   35.273835] flags: 0xfffff8000008000(head)
[   35.282007] raw: 0fffff8000008000 dead000000000100 dead000000000200 0000000000000000
[   35.282010] raw: 0000000000000000 0000000000000000 00000001ffffffff 0000000000000000
[   35.282012] page dumped because: kasan: bad access detected

[   35.282014] Memory state around the buggy address:
[   35.282017]  ffff803b66515700: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
[   35.282019]  ffff803b66515780: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
[   35.282021] >ffff803b66515800: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
[   35.282022]                             ^
[   35.282024]  ffff803b66515880: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
[   35.282026]  ffff803b66515900: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
[   35.282028] ==================================================================
[   35.282029] Disabling lock debugging due to kernel taint
[   35.282747] hclge driver initialization finished.

Fixes: 67bf2541 ("net: hns3: Fixes the back pressure setting when sriov is enabled")
Signed-off-by: default avatarYunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      e8ccbb7d
    • David S. Miller's avatar
      Merge branch 'net-bql-better-deal-with-GSO' · cb53fd54
      David S. Miller authored 
      Eric Dumazet says:

====================
net: bql: better deal with GSO

While BQL bulk dequeue works well for TSO packets, it is
not very efficient as soon as GSO is involved.

On a GSO only workload (UDP or TCP), this patch series
can save about 8 % of cpu cycles on a 40Gbit mlx4 NIC,
by keeping optimal batching, and avoiding expensive
doorbells, qdisc requeues and reschedules.

This patch series :

- Add __netdev_tx_sent_queue() so that drivers
  can implement efficient BQL and xmit_more support.

- Implement a work around in dev_hard_start_xmit()
  for drivers not using __netdev_tx_sent_queue()

- changes mlx4 to use __netdev_tx_sent_queue()

v2: Tariq and Willem feedback addressed.
    added __netdev_tx_sent_queue() (Willem suggestion)
====================
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      cb53fd54
    • Eric Dumazet's avatar
      net/mlx4_en: use __netdev_tx_sent_queue() · c2973444
      Eric Dumazet authored 
      doorbell only depends on xmit_more and netif_tx_queue_stopped()

Using __netdev_tx_sent_queue() avoids messing with BQL stop flag,
and is more generic.

This patch increases performance on GSO workload by keeping
doorbells to the minimum required.
Signed-off-by: default avatarEric Dumazet <edumazet@google.com>
Cc: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: default avatarTariq Toukan <tariqt@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      c2973444
    • Eric Dumazet's avatar
      net: do not abort bulk send on BQL status · fe60faa5
      Eric Dumazet authored 
      Before calling dev_hard_start_xmit(), upper layers tried
to cook optimal skb list based on BQL budget.

Problem is that GSO packets can end up comsuming more than
the BQL budget.

Breaking the loop is not useful, since requeued packets
are ahead of any packets still in the qdisc.

It is also more expensive, since next TX completion will
push these packets later, while skbs are not in cpu caches.

It is also a behavior difference with TSO packets, that can
break the BQL limit by a large amount.

Note that drivers should use __netdev_tx_sent_queue()
in order to have optimal xmit_more support, and avoid
useless atomic operations as shown in the following patch.
Signed-off-by: default avatarEric Dumazet <edumazet@google.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      fe60faa5
    • Eric Dumazet's avatar
      net: bql: add __netdev_tx_sent_queue() · 3e59020a
      Eric Dumazet authored 
      When qdisc_run() tries to use BQL budget to bulk-dequeue a batch
of packets, GSO can later transform this list in another list
of skbs, and each skb is sent to device ndo_start_xmit(),
one at a time, with skb->xmit_more being set to one but
for last skb.

Problem is that very often, BQL limit is hit in the middle of
the packet train, forcing dev_hard_start_xmit() to stop the
bulk send and requeue the end of the list.

BQL role is to avoid head of line blocking, making sure
a qdisc can deliver high priority packets before low priority ones.

But there is no way requeued packets can be bypassed by fresh
packets in the qdisc.

Aborting the bulk send increases TX softirqs, and hot cache
lines (after skb_segment()) are wasted.

Note that for TSO packets, we never split a packet in the middle
because of BQL limit being hit.

Drivers should be able to update BQL counters without
flipping/caring about BQL status, if the current skb
has xmit_more set.

Upper layers are ultimately responsible to stop sending another
packet train when BQL limit is hit.

Code template in a driver might look like the following :

	send_doorbell = __netdev_tx_sent_queue(tx_queue, nr_bytes, skb->xmit_more);

Note that __netdev_tx_sent_queue() use is not mandatory,
since following patch will change dev_hard_start_xmit()
to not care about BQL status.

But it is highly recommended so that xmit_more full benefits
can be reached (less doorbells sent, and less atomic operations as well)
Signed-off-by: default avatarEric Dumazet <edumazet@google.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      3e59020a
    • David S. Miller's avatar
      Merge branch 's390-qeth-fixes' · b8a5d06a
      David S. Miller authored 
      Julian Wiedmann says:

====================
s390/qeth: fixes 2018-11-02

please apply one round of qeth fixes for -net.

Patch 1 is rather large and removes a use-after-free hazard from many of our
debug trace entries.
Patch 2 is yet another fix-up for the L3 subdriver's new IP address management
code.
Patch 3 and 4 resolve some fallout from the recent changes wrt how/when qeth
allocates its net_device.
Patch 5 makes sure we don't set reserved bits when building HW commands from
user-provided data.
And finally, patch 6 allows ethtool to play nice with new HW.
====================
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      b8a5d06a
    • Julian Wiedmann's avatar
      s390/qeth: report 25Gbit link speed · 54e049c2
      Julian Wiedmann authored 
      This adds the various identifiers for 25Gbit cards, and wires them up
into sysfs and ethtool.
Signed-off-by: default avatarJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      54e049c2
    • Julian Wiedmann's avatar
      s390/qeth: sanitize ARP requests · 125d7d30
      Julian Wiedmann authored 
      The ARP_{ADD,REMOVE}_ENTRY cmd structs contain reserved fields.
Introduce a common helper that doesn't raw-copy the user-provided data
into the cmd, but only sets those fields that are strictly needed for
the command.

This also sets the correct command length for ARP_REMOVE_ENTRY.
Signed-off-by: default avatarJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      125d7d30
    • Julian Wiedmann's avatar
      s390/qeth: fix initial operstate · 9fae5c3b
      Julian Wiedmann authored 
      Setting the carrier 'on' for an unregistered netdevice doesn't update
its operstate. Fix this by delaying the update until the netdevice has
been registered.

Fixes: 91cc98f5 ("s390/qeth: remove duplicated carrier state tracking")
Signed-off-by: default avatarJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      9fae5c3b
    • Julian Wiedmann's avatar
      s390/qeth: unregister netdevice only when registered · 30356d08
      Julian Wiedmann authored 
      qeth only registers its netdevice when the qeth device is first set
online. Thus a device that has never been set online will trigger
a WARN ("network todo 'hsi%d' but state 0") in unregister_netdev() when
removed.

Fix this by protecting the unregister step, just like we already protect
against repeated registering of the netdevice.

Fixes: d3d1b205 ("s390/qeth: allocate netdevice early")
Reported-by: default avatarKarsten Graul <kgraul@linux.ibm.com>
Signed-off-by: default avatarJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      30356d08
    • Julian Wiedmann's avatar
      s390/qeth: fix HiperSockets sniffer · bd74a7f9
      Julian Wiedmann authored 
      Sniffing mode for L3 HiperSockets requires that no IP addresses are
registered with the HW. The preferred way to achieve this is for
userspace to delete all the IPs on the interface. But qeth is expected
to also tolerate a configuration where that is not the case, by skipping
the IP registration when in sniffer mode.
Since commit 5f78e29c ("qeth: optimize IP handling in rx_mode callback")
reworked the IP registration logic in the L3 subdriver, this no longer
works. When the qeth device is set online, qeth_l3_recover_ip() now
unconditionally registers all unicast addresses from our internal
IP table.

While we could fix this particular problem by skipping
qeth_l3_recover_ip() on a sniffer device, the more future-proof change
is to skip the IP address registration at the lowest level. This way we
a) catch any future code path that attempts to register an IP address
   without considering the sniffer scenario, and
b) continue to build up our internal IP table, so that if sniffer mode
   is switched off later we can operate just like normal.

Fixes: 5f78e29c ("qeth: optimize IP handling in rx_mode callback")
Signed-off-by: default avatarJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      bd74a7f9
    • Julian Wiedmann's avatar
      s390/qeth: sanitize strings in debug messages · e19e5be8
      Julian Wiedmann authored 
      As Documentation/s390/s390dbf.txt states quite clearly, using any
pointer in sprinf-formatted s390dbf debug entries is dangerous.
The pointers are dereferenced whenever the trace file is read from.
So if the referenced data has a shorter life-time than the trace file,
any read operation can result in a use-after-free.

So rip out all hazardous use of indirect data, and replace any usage of
dev_name() and such by the Bus ID number.
Signed-off-by: default avatarJulian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      e19e5be8
    • David S. Miller's avatar
      Merge branch 'net-timeout-fixes-for-GENET-and-SYSTEMPORT' · 265ad063
      David S. Miller authored 
      Florian Fainelli says:

====================
net: timeout fixes for GENET and SYSTEMPORT

This patch series fixes occasional transmit timeout around the time
the system goes into suspend. GENET and SYSTEMPORT have nearly the same
logic in that regard and were both affected in the same way.

Please queue up for stable, thanks!
====================
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      265ad063
    • Florian Fainelli's avatar
      net: systemport: Protect stop from timeout · 7cb6a2a2
      Florian Fainelli authored 
      A timing hazard exists when the network interface is stopped that
allows a watchdog timeout to be processed by a separate core in
parallel. This creates the potential for the timeout handler to
wake the queues while the driver is shutting down, or access
registers after their clocks have been removed.

The more common case is that the watchdog timeout will produce a
warning message which doesn't lead to a crash. The chances of this
are greatly increased by the fact that bcm_sysport_netif_stop stops
the transmit queues which can easily precipitate a watchdog time-
out because of stale trans_start data in the queues.

This commit corrects the behavior by ensuring that the watchdog
timeout is disabled before enterring bcm_sysport_netif_stop. There
are currently only two users of the bcm_sysport_netif_stop function:
close and suspend.

The close case already handles the issue by exiting the RUNNING
state before invoking the driver close service.

The suspend case now performs the netif_device_detach to exit the
PRESENT state before the call to bcm_sysport_netif_stop rather than
after it.

These behaviors prevent any future scheduling of the driver timeout
service during the window. The netif_tx_stop_all_queues function
in bcm_sysport_netif_stop is replaced with netif_tx_disable to ensure
synchronization with any transmit or timeout threads that may
already be executing on other cores.

For symmetry, the netif_device_attach call upon resume is moved to
after the call to bcm_sysport_netif_start. Since it wakes the transmit
queues it is not necessary to invoke netif_tx_start_all_queues from
bcm_sysport_netif_start so it is moved into the driver open service.

Fixes: 40755a0f ("net: systemport: add suspend and resume support")
Fixes: 80105bef ("net: systemport: add Broadcom SYSTEMPORT Ethernet MAC driver")
Signed-off-by: default avatarFlorian Fainelli <f.fainelli@gmail.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      7cb6a2a2
    • Doug Berger's avatar
      net: bcmgenet: protect stop from timeout · 09e805d2
      Doug Berger authored 
      A timing hazard exists when the network interface is stopped that
allows a watchdog timeout to be processed by a separate core in
parallel. This creates the potential for the timeout handler to
wake the queues while the driver is shutting down, or access
registers after their clocks have been removed.

The more common case is that the watchdog timeout will produce a
warning message which doesn't lead to a crash. The chances of this
are greatly increased by the fact that bcmgenet_netif_stop stops
the transmit queues which can easily precipitate a watchdog time-
out because of stale trans_start data in the queues.

This commit corrects the behavior by ensuring that the watchdog
timeout is disabled before enterring bcmgenet_netif_stop. There
are currently only two users of the bcmgenet_netif_stop function:
close and suspend.

The close case already handles the issue by exiting the RUNNING
state before invoking the driver close service.

The suspend case now performs the netif_device_detach to exit the
PRESENT state before the call to bcmgenet_netif_stop rather than
after it.

These behaviors prevent any future scheduling of the driver timeout
service during the window. The netif_tx_stop_all_queues function
in bcmgenet_netif_stop is replaced with netif_tx_disable to ensure
synchronization with any transmit or timeout threads that may
already be executing on other cores.

For symmetry, the netif_device_attach call upon resume is moved to
after the call to bcmgenet_netif_start. Since it wakes the transmit
queues it is not necessary to invoke netif_tx_start_all_queues from
bcmgenet_netif_start so it is moved into the driver open service.

Fixes: 1c1008c7 ("net: bcmgenet: add main driver file")
Signed-off-by: default avatarDoug Berger <opendmb@gmail.com>
Signed-off-by: default avatarFlorian Fainelli <f.fainelli@gmail.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      09e805d2
    • David Howells's avatar
      rxrpc: Fix lockup due to no error backoff after ack transmit error · c7e86acf
      David Howells authored 
      If the network becomes (partially) unavailable, say by disabling IPv6, the
background ACK transmission routine can get itself into a tizzy by
proposing immediate ACK retransmission.  Since we're in the call event
processor, that happens immediately without returning to the workqueue
manager.

The condition should clear after a while when either the network comes back
or the call times out.

Fix this by:

 (1) When re-proposing an ACK on failed Tx, don't schedule it immediately.
     This will allow a certain amount of time to elapse before we try
     again.

 (2) Enforce a return to the workqueue manager after a certain number of
     iterations of the call processing loop.

 (3) Add a backoff delay that increases the delay on deferred ACKs by a
     jiffy per failed transmission to a limit of HZ.  The backoff delay is
     cleared on a successful return from kernel_sendmsg().

 (4) Cancel calls immediately if the opening sendmsg fails.  The layer
     above can arrange retransmission or rotate to another server.

Fixes: 248f219c ("rxrpc: Rewrite the data and ack handling code")
Signed-off-by: default avatarDavid Howells <dhowells@redhat.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      c7e86acf
    • Tristram Ha's avatar
      net: dsa: microchip: initialize mutex before use · 284fb78e
      Tristram Ha authored 
      Initialize mutex before use.  Avoid kernel complaint when
CONFIG_DEBUG_LOCK_ALLOC is enabled.

Fixes: b987e98e ("dsa: add DSA switch driver for Microchip KSZ9477")
Signed-off-by: default avatarTristram Ha <Tristram.Ha@microchip.com>
Reviewed-by: default avatarPavel Machek <pavel@ucw.cz>
Reviewed-by: default avatarAndrew Lunn <andrew@lunn.ch>
Reviewed-by: default avatarFlorian Fainelli <f.fainelli@gmail.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      284fb78e
    • Jeff Barnhill's avatar
      net/ipv6: Add anycast addresses to a global hashtable · 2384d025
      Jeff Barnhill authored 
      icmp6_send() function is expensive on systems with a large number of
interfaces. Every time it’s called, it has to verify that the source
address does not correspond to an existing anycast address by looping
through every device and every anycast address on the device.  This can
result in significant delays for a CPU when there are a large number of
neighbors and ND timers are frequently timing out and calling
neigh_invalidate().

Add anycast addresses to a global hashtable to allow quick searching for
matching anycast addresses.  This is based on inet6_addr_lst in addrconf.c.
Signed-off-by: default avatarJeff Barnhill <0xeffeff@gmail.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      2384d025
    • Frieder Schrempf's avatar
      usbnet: smsc95xx: disable carrier check while suspending · 7b900ead
      Frieder Schrempf authored 
      We need to make sure, that the carrier check polling is disabled
while suspending. Otherwise we can end up with usbnet_read_cmd()
being issued when only usbnet_read_cmd_nopm() is allowed. If this
happens, read operations lock up.

Fixes: d69d1694 ("usbnet: smsc95xx: fix link detection for disabled autonegotiation")
Signed-off-by: default avatarFrieder Schrempf <frieder.schrempf@kontron.de>
Reviewed-by: default avatarRaghuram Chary J <RaghuramChary.Jallipalli@microchip.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      7b900ead
    • Mathieu Malaterre's avatar
      net: document skb parameter in function 'skb_gso_size_check' · 49682bfa
      Mathieu Malaterre authored 
      Remove kernel-doc warning:

  net/core/skbuff.c:4953: warning: Function parameter or member 'skb' not described in 'skb_gso_size_check'
Signed-off-by: default avatarMathieu Malaterre <malat@debian.org>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      49682bfa
  9. 02 Nov, 2018 5 commits
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7