openwrt/staging/blogic.git
4 years agos390/qeth: don't warn for napi with 0 budget
Julian Wiedmann [Thu, 20 Feb 2020 14:54:55 +0000 (15:54 +0100)]
s390/qeth: don't warn for napi with 0 budget

Calling napi->poll() with 0 budget is a legitimate use by netpoll.

Fixes: a1c3ed4c9ca0 ("qeth: NAPI support for l2 and l3 discipline")
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agos390/qeth: vnicc Fix EOPNOTSUPP precedence
Alexandra Winter [Thu, 20 Feb 2020 14:54:54 +0000 (15:54 +0100)]
s390/qeth: vnicc Fix EOPNOTSUPP precedence

When getting or setting VNICC parameters, the error code EOPNOTSUPP
should have precedence over EBUSY.

EBUSY is used because vnicc feature and bridgeport feature are mutually
exclusive, which is a temporary condition.
Whereas EOPNOTSUPP indicates that the HW does not support all or parts of
the vnicc feature.
This issue causes the vnicc sysfs params to show 'blocked by bridgeport'
for HW that does not support VNICC at all.

Fixes: caa1f0b10d18 ("s390/qeth: add VNICC enable/disable support")
Signed-off-by: Alexandra Winter <wintera@linux.ibm.com>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoopenvswitch: Distribute switch variables for initialization
Kees Cook [Thu, 20 Feb 2020 06:23:09 +0000 (22:23 -0800)]
openvswitch: Distribute switch variables for initialization

Variables declared in a switch statement before any case statements
cannot be automatically initialized with compiler instrumentation (as
they are not part of any execution flow). With GCC's proposed automatic
stack variable initialization feature, this triggers a warning (and they
don't get initialized). Clang's automatic stack variable initialization
(via CONFIG_INIT_STACK_ALL=y) doesn't throw a warning, but it also
doesn't initialize such variables[1]. Note that these warnings (or silent
skipping) happen before the dead-store elimination optimization phase,
so even when the automatic initializations are later elided in favor of
direct initializations, the warnings remain.

To avoid these problems, move such variables into the "case" where
they're used or lift them up into the main function body.

net/openvswitch/flow_netlink.c: In function ‘validate_set’:
net/openvswitch/flow_netlink.c:2711:29: warning: statement will never be executed [-Wswitch-unreachable]
 2711 |  const struct ovs_key_ipv4 *ipv4_key;
      |                             ^~~~~~~~

[1] https://bugs.llvm.org/show_bug.cgi?id=44916

Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ip6_gre: Distribute switch variables for initialization
Kees Cook [Thu, 20 Feb 2020 06:23:07 +0000 (22:23 -0800)]
net: ip6_gre: Distribute switch variables for initialization

Variables declared in a switch statement before any case statements
cannot be automatically initialized with compiler instrumentation (as
they are not part of any execution flow). With GCC's proposed automatic
stack variable initialization feature, this triggers a warning (and they
don't get initialized). Clang's automatic stack variable initialization
(via CONFIG_INIT_STACK_ALL=y) doesn't throw a warning, but it also
doesn't initialize such variables[1]. Note that these warnings (or silent
skipping) happen before the dead-store elimination optimization phase,
so even when the automatic initializations are later elided in favor of
direct initializations, the warnings remain.

To avoid these problems, move such variables into the "case" where
they're used or lift them up into the main function body.

net/ipv6/ip6_gre.c: In function ‘ip6gre_err’:
net/ipv6/ip6_gre.c:440:32: warning: statement will never be executed [-Wswitch-unreachable]
  440 |   struct ipv6_tlv_tnl_enc_lim *tel;
      |                                ^~~

net/ipv6/ip6_tunnel.c: In function ‘ip6_tnl_err’:
net/ipv6/ip6_tunnel.c:520:32: warning: statement will never be executed [-Wswitch-unreachable]
  520 |   struct ipv6_tlv_tnl_enc_lim *tel;
      |                                ^~~

[1] https://bugs.llvm.org/show_bug.cgi?id=44916

Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: core: Distribute switch variables for initialization
Kees Cook [Thu, 20 Feb 2020 06:23:04 +0000 (22:23 -0800)]
net: core: Distribute switch variables for initialization

Variables declared in a switch statement before any case statements
cannot be automatically initialized with compiler instrumentation (as
they are not part of any execution flow). With GCC's proposed automatic
stack variable initialization feature, this triggers a warning (and they
don't get initialized). Clang's automatic stack variable initialization
(via CONFIG_INIT_STACK_ALL=y) doesn't throw a warning, but it also
doesn't initialize such variables[1]. Note that these warnings (or silent
skipping) happen before the dead-store elimination optimization phase,
so even when the automatic initializations are later elided in favor of
direct initializations, the warnings remain.

To avoid these problems, move such variables into the "case" where
they're used or lift them up into the main function body.

net/core/skbuff.c: In function ‘skb_checksum_setup_ip’:
net/core/skbuff.c:4809:7: warning: statement will never be executed [-Wswitch-unreachable]
 4809 |   int err;
      |       ^~~

[1] https://bugs.llvm.org/show_bug.cgi?id=44916

Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
David S. Miller [Thu, 20 Feb 2020 00:42:35 +0000 (16:42 -0800)]
Merge git://git./pub/scm/linux/kernel/git/bpf/bpf

Alexei Starovoitov says:

====================
pull-request: bpf 2020-02-19

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

We've added 10 non-merge commits during the last 10 day(s) which contain
a total of 10 files changed, 93 insertions(+), 31 deletions(-).

The main changes are:

1) batched bpf hashtab fixes from Brian and Yonghong.

2) various selftests and libbpf fixes.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net...
David S. Miller [Thu, 20 Feb 2020 00:38:16 +0000 (16:38 -0800)]
Merge branch '100GbE' of git://git./linux/kernel/git/jkirsher/net-queue

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2020-02-19

This series contains fixes to the ice driver.

Brett fixes an issue where if a user sets an odd [tx|rx]-usecs value
through ethtool, the request is denied because the hardware is set to
have an ITR with 2us granularity.  Also fix an issue where the VF has
not been completely removed/reset after being unbound from the host
driver, so resolve this by waiting for the VF remove/reset process to
happen before checking if the VF is disabled.

Michal fixes an issue, where when the user changes flow control via
ethtool, the OS is told the link is going down when that may not be the
case.  Before the fix, the only way to get out of this state was to take
the interface down and up again.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoudp: rehash on disconnect
Willem de Bruijn [Wed, 19 Feb 2020 19:16:32 +0000 (14:16 -0500)]
udp: rehash on disconnect

As of the below commit, udp sockets bound to a specific address can
coexist with one bound to the any addr for the same port.

The commit also phased out the use of socket hashing based only on
port (hslot), in favor of always hashing on {addr, port} (hslot2).

The change broke the following behavior with disconnect (AF_UNSPEC):

    server binds to 0.0.0.0:1337
    server connects to 127.0.0.1:80
    server disconnects
    client connects to 127.0.0.1:1337
    client sends "hello"
    server reads "hello" // times out, packet did not find sk

On connect the server acquires a specific source addr suitable for
routing to its destination. On disconnect it reverts to the any addr.

The connect call triggers a rehash to a different hslot2. On
disconnect, add the same to return to the original hslot2.

Skip this step if the socket is going to be unhashed completely.

Fixes: 4cdeeee9252a ("net: udp: prefer listeners bound to an address")
Reported-by: Pavel Roskin <plroskin@gmail.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/tls: Fix to avoid gettig invalid tls record
Rohit Maheshwari [Wed, 19 Feb 2020 04:10:22 +0000 (09:40 +0530)]
net/tls: Fix to avoid gettig invalid tls record

Current code doesn't check if tcp sequence number is starting from (/after)
1st record's start sequnce number. It only checks if seq number is before
1st record's end sequnce number. This problem will always be a possibility
in re-transmit case. If a record which belongs to a requested seq number is
already deleted, tls_get_record will start looking into list and as per the
check it will look if seq number is before the end seq of 1st record, which
will always be true and will return 1st record always, it should in fact
return NULL.
As part of the fix, start looking each record only if the sequence number
lies in the list else return NULL.
There is one more check added, driver look for the start marker record to
handle tcp packets which are before the tls offload start sequence number,
hence return 1st record if the record is tls start marker and seq number is
before the 1st record's starting sequence number.

Fixes: e8f69799810c ("net/tls: Add generic NIC offload infrastructure")
Signed-off-by: Rohit Maheshwari <rohitm@chelsio.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobpf: Fix a potential deadlock with bpf_map_do_batch
Yonghong Song [Wed, 19 Feb 2020 23:47:57 +0000 (15:47 -0800)]
bpf: Fix a potential deadlock with bpf_map_do_batch

Commit 057996380a42 ("bpf: Add batch ops to all htab bpf map")
added lookup_and_delete batch operation for hash table.
The current implementation has bpf_lru_push_free() inside
the bucket lock, which may cause a deadlock.

syzbot reports:
   -> #2 (&htab->buckets[i].lock#2){....}:
       __raw_spin_lock_irqsave include/linux/spinlock_api_smp.h:110 [inline]
       _raw_spin_lock_irqsave+0x95/0xcd kernel/locking/spinlock.c:159
       htab_lru_map_delete_node+0xce/0x2f0 kernel/bpf/hashtab.c:593
       __bpf_lru_list_shrink_inactive kernel/bpf/bpf_lru_list.c:220 [inline]
       __bpf_lru_list_shrink+0xf9/0x470 kernel/bpf/bpf_lru_list.c:266
       bpf_lru_list_pop_free_to_local kernel/bpf/bpf_lru_list.c:340 [inline]
       bpf_common_lru_pop_free kernel/bpf/bpf_lru_list.c:447 [inline]
       bpf_lru_pop_free+0x87c/0x1670 kernel/bpf/bpf_lru_list.c:499
       prealloc_lru_pop+0x2c/0xa0 kernel/bpf/hashtab.c:132
       __htab_lru_percpu_map_update_elem+0x67e/0xa90 kernel/bpf/hashtab.c:1069
       bpf_percpu_hash_update+0x16e/0x210 kernel/bpf/hashtab.c:1585
       bpf_map_update_value.isra.0+0x2d7/0x8e0 kernel/bpf/syscall.c:181
       generic_map_update_batch+0x41f/0x610 kernel/bpf/syscall.c:1319
       bpf_map_do_batch+0x3f5/0x510 kernel/bpf/syscall.c:3348
       __do_sys_bpf+0x9b7/0x41e0 kernel/bpf/syscall.c:3460
       __se_sys_bpf kernel/bpf/syscall.c:3355 [inline]
       __x64_sys_bpf+0x73/0xb0 kernel/bpf/syscall.c:3355
       do_syscall_64+0xfa/0x790 arch/x86/entry/common.c:294
       entry_SYSCALL_64_after_hwframe+0x49/0xbe

   -> #0 (&loc_l->lock){....}:
       check_prev_add kernel/locking/lockdep.c:2475 [inline]
       check_prevs_add kernel/locking/lockdep.c:2580 [inline]
       validate_chain kernel/locking/lockdep.c:2970 [inline]
       __lock_acquire+0x2596/0x4a00 kernel/locking/lockdep.c:3954
       lock_acquire+0x190/0x410 kernel/locking/lockdep.c:4484
       __raw_spin_lock_irqsave include/linux/spinlock_api_smp.h:110 [inline]
       _raw_spin_lock_irqsave+0x95/0xcd kernel/locking/spinlock.c:159
       bpf_common_lru_push_free kernel/bpf/bpf_lru_list.c:516 [inline]
       bpf_lru_push_free+0x250/0x5b0 kernel/bpf/bpf_lru_list.c:555
       __htab_map_lookup_and_delete_batch+0x8d4/0x1540 kernel/bpf/hashtab.c:1374
       htab_lru_map_lookup_and_delete_batch+0x34/0x40 kernel/bpf/hashtab.c:1491
       bpf_map_do_batch+0x3f5/0x510 kernel/bpf/syscall.c:3348
       __do_sys_bpf+0x1f7d/0x41e0 kernel/bpf/syscall.c:3456
       __se_sys_bpf kernel/bpf/syscall.c:3355 [inline]
       __x64_sys_bpf+0x73/0xb0 kernel/bpf/syscall.c:3355
       do_syscall_64+0xfa/0x790 arch/x86/entry/common.c:294
       entry_SYSCALL_64_after_hwframe+0x49/0xbe

    Possible unsafe locking scenario:

          CPU0                    CPU2
          ----                    ----
     lock(&htab->buckets[i].lock#2);
                                  lock(&l->lock);
                                  lock(&htab->buckets[i].lock#2);
     lock(&loc_l->lock);

    *** DEADLOCK ***

To fix the issue, for htab_lru_map_lookup_and_delete_batch() in CPU0,
let us do bpf_lru_push_free() out of the htab bucket lock. This can
avoid the above deadlock scenario.

Fixes: 057996380a42 ("bpf: Add batch ops to all htab bpf map")
Reported-by: syzbot+a38ff3d9356388f2fb83@syzkaller.appspotmail.com
Reported-by: syzbot+122b5421d14e68f29cd1@syzkaller.appspotmail.com
Suggested-by: Hillf Danton <hdanton@sina.com>
Suggested-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Acked-by: Brian Vazquez <brianvv@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20200219234757.3544014-1-yhs@fb.com
4 years agobpf: Do not grab the bucket spinlock by default on htab batch ops
Brian Vazquez [Tue, 18 Feb 2020 17:25:52 +0000 (09:25 -0800)]
bpf: Do not grab the bucket spinlock by default on htab batch ops

Grabbing the spinlock for every bucket even if it's empty, was causing
significant perfomance cost when traversing htab maps that have only a
few entries. This patch addresses the issue by checking first the
bucket_cnt, if the bucket has some entries then we go and grab the
spinlock and proceed with the batching.

Tested with a htab of size 50K and different value of populated entries.

Before:
  Benchmark             Time(ns)        CPU(ns)
  ---------------------------------------------
  BM_DumpHashMap/1       2759655        2752033
  BM_DumpHashMap/10      2933722        2930825
  BM_DumpHashMap/200     3171680        3170265
  BM_DumpHashMap/500     3639607        3635511
  BM_DumpHashMap/1000    4369008        4364981
  BM_DumpHashMap/5k     11171919       11134028
  BM_DumpHashMap/20k    69150080       69033496
  BM_DumpHashMap/39k   190501036      190226162

After:
  Benchmark             Time(ns)        CPU(ns)
  ---------------------------------------------
  BM_DumpHashMap/1        202707         200109
  BM_DumpHashMap/10       213441         210569
  BM_DumpHashMap/200      478641         472350
  BM_DumpHashMap/500      980061         967102
  BM_DumpHashMap/1000    1863835        1839575
  BM_DumpHashMap/5k      8961836        8902540
  BM_DumpHashMap/20k    69761497       69322756
  BM_DumpHashMap/39k   187437830      186551111

Fixes: 057996380a42 ("bpf: Add batch ops to all htab bpf map")
Signed-off-by: Brian Vazquez <brianvv@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200218172552.215077-1-brianvv@google.com
4 years agoice: Wait for VF to be reset/ready before configuration
Brett Creeley [Tue, 18 Feb 2020 21:22:06 +0000 (13:22 -0800)]
ice: Wait for VF to be reset/ready before configuration

The configuration/command below is failing when the VF in the xml
file is already bound to the host iavf driver.

pci_0000_af_0_0.xml:

<interface type='hostdev' managed='yes'>
<source>
<address type='pci' domain='0x0000' bus='0xaf' slot='0x0' function='0x0'/>
</source>
<mac address='00:de:ad:00:11:01'/>
</interface>

> virsh attach-device domain_name pci_0000_af_0_0.xml
error: Failed to attach device from pci_0000_af_0_0.xml
error: Cannot set interface MAC/vlanid to 00:de:ad:00:11:01/0 for
ifname ens1f1 vf 0: Device or resource busy

This is failing because the VF has not been completely removed/reset
after being unbound (via the virsh command above) from the host iavf
driver and ice_set_vf_mac() checks if the VF is disabled before waiting
for the reset to finish.

Fix this by waiting for the VF remove/reset process to happen before
checking if the VF is disabled. Also, since many functions for VF
administration on the PF were more or less calling the same 3 functions
(ice_wait_on_vf_reset(), ice_is_vf_disabled(), and ice_check_vf_init())
move these into the helper function ice_check_vf_ready_for_cfg(). Then
call this function in any flow that attempts to configure/query a VF
from the PF.

Lastly, increase the maximum wait time in ice_wait_on_vf_reset() to
800ms, and modify/add the #define(s) that determine the wait time.
This was done for robustness because in rare/stress cases VF removal can
take a max of ~800ms and previously the wait was a max of ~300ms.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: Don't tell the OS that link is going down
Michal Swiatkowski [Thu, 13 Feb 2020 21:31:24 +0000 (13:31 -0800)]
ice: Don't tell the OS that link is going down

Remove code that tell the OS that link is going down when user
change flow control via ethtool. When link is up it isn't certain
that link goes down after 0x0605 aq command. If link doesn't go
down, OS thinks that link is down, but physical link is up. To
reset this state user have to take interface down and up.

If link goes down after 0x0605 command, FW send information
about that and after that driver tells the OS that the link goes
down. So this code in ethtool is unnecessary.

Signed-off-by: Michal Swiatkowski <michal.swiatkowski@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agoice: Don't reject odd values of usecs set by user
Brett Creeley [Thu, 13 Feb 2020 21:31:23 +0000 (13:31 -0800)]
ice: Don't reject odd values of usecs set by user

Currently if a user sets an odd [tx|rx]-usecs value through ethtool,
the request is denied because the hardware is set to have an ITR
granularity of 2us. This caused poor customer experience. Fix this by
aligning to a register allowed value, which results in rounding down.
Also, print a once per ring container type message to be clear about
our intentions.

Also, change the ITR_TO_REG define to be the bitwise and of the ITR
setting and the ICE_ITR_MASK. This makes the purpose of ITR_TO_REG more
obvious.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
4 years agobridge: br_stp: Use built-in RCU list checking
Madhuparna Bhowmik [Wed, 19 Feb 2020 15:17:46 +0000 (20:47 +0530)]
bridge: br_stp: Use built-in RCU list checking

list_for_each_entry_rcu() has built-in RCU and lock checking.

Pass cond argument to list_for_each_entry_rcu() to silence
false lockdep warning when CONFIG_PROVE_RCU_LIST is enabled
by default.

Signed-off-by: Madhuparna Bhowmik <madhuparnabhowmik10@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfc: pn544: Fix occasional HW initialization failure
Dmitry Osipenko [Wed, 19 Feb 2020 15:01:22 +0000 (18:01 +0300)]
nfc: pn544: Fix occasional HW initialization failure

The PN544 driver checks the "enable" polarity during of driver's probe and
it's doing that by turning ON and OFF NFC with different polarities until
enabling succeeds. It takes some time for the hardware to power-down, and
thus, to deassert the IRQ that is raised by turning ON the hardware.
Since the delay after last power-down of the polarity-checking process is
missed in the code, the interrupt may trigger immediately after installing
the IRQ handler (right after the checking is done), which results in IRQ
handler trying to touch the disabled HW and ends with marking NFC as
'DEAD' during of the driver's probe:

  pn544_hci_i2c 1-002a: NFC: nfc_en polarity : active high
  pn544_hci_i2c 1-002a: NFC: invalid len byte
  shdlc: llc_shdlc_recv_frame: NULL Frame -> link is dead

This patch fixes the occasional NFC initialization failure on Nexus 7
device.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hsr: Pass lockdep expression to RCU lists
Amol Grover [Wed, 19 Feb 2020 10:00:11 +0000 (15:30 +0530)]
net: hsr: Pass lockdep expression to RCU lists

node_db is traversed using list_for_each_entry_rcu
outside an RCU read-side critical section but under the protection
of hsr->list_lock.

Hence, add corresponding lockdep expression to silence false-positive
warnings, and harden RCU lists.

Signed-off-by: Amol Grover <frextrite@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'mlx5-fixes-2020-02-18' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Wed, 19 Feb 2020 18:31:26 +0000 (10:31 -0800)]
Merge tag 'mlx5-fixes-2020-02-18' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
Mellanox, mlx5 fixes 2020-02-18

This series introduces some fixes to mlx5 driver.

Please pull and let me know if there is any problem.

For -stable v5.3
 ('net/mlx5: Fix sleep while atomic in mlx5_eswitch_get_vepa')

For -stable v5.4
 ('net/mlx5: DR, Fix matching on vport gvmi')
 ('net/mlx5e: Fix crash in recovery flow without devlink reporter')

For -stable v5.5
 ('net/mlx5e: Reset RQ doorbell counter before moving RQ state from RST to RDY')
 ('net/mlx5e: Don't clear the whole vf config when switching modes')
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlx5: DR, Handle reformat capability over sw-steering tables
Erez Shitrit [Tue, 14 Jan 2020 07:27:27 +0000 (09:27 +0200)]
net/mlx5: DR, Handle reformat capability over sw-steering tables

On flow table creation, send the relevant flags according to what the FW
currently supports.
When FW doesn't support reformat option over SW-steering managed table,
the driver shouldn't pass this.

Fixes: 988fd6b32d07 ("net/mlx5: DR, Pass table flags at creation to lower layer")
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Fix lowest FDB pool size
Paul Blakey [Tue, 31 Dec 2019 15:04:15 +0000 (17:04 +0200)]
net/mlx5: Fix lowest FDB pool size

The pool sizes represent the pool sizes in the fw. when we request
a pool size from fw, it will return the next possible group.
We track how many pools the fw has left and start requesting groups
from the big to the small.
When we start request 4k group, which doesn't exists in fw, fw
wants to allocate the next possible size, 64k, but will fail since
its exhausted. The correct smallest pool size in fw is 128 and not 4k.

Fixes: 39ac237ce009 ("net/mlx5: E-Switch, Refactor chains and priorities")
Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Don't clear the whole vf config when switching modes
Dmytro Linkin [Wed, 12 Feb 2020 09:32:39 +0000 (11:32 +0200)]
net/mlx5e: Don't clear the whole vf config when switching modes

There is no need to reset all vf config (except link state) between
legacy and switchdev modes changes.
Also, set link state to AUTO, when legacy enabled.

Fixes: 3b83b6c2e024 ("net/mlx5e: Clear VF config when switching modes")
Signed-off-by: Dmytro Linkin <dmitrolin@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: DR, Fix matching on vport gvmi
Hamdan Igbaria [Wed, 5 Feb 2020 12:31:12 +0000 (14:31 +0200)]
net/mlx5: DR, Fix matching on vport gvmi

Set vport gvmi in the tag, only when source gvmi is set in the bit mask.

Fixes: 26d688e3 ("net/mlx5: DR, Add Steering entry (STE) utilities")
Signed-off-by: Hamdan Igbaria <hamdani@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Fix crash in recovery flow without devlink reporter
Aya Levin [Wed, 12 Feb 2020 13:17:25 +0000 (15:17 +0200)]
net/mlx5e: Fix crash in recovery flow without devlink reporter

When health reporters are not supported, recovery function is invoked
directly, not via devlink health reporters.

In this direct flow, the recover function input parameter was passed
incorrectly and is causing a kernel oops. This patch is fixing the input
parameter.

Following call trace is observed on rx error health reporting.

Internal error: Oops: 96000007 [#1] PREEMPT SMP
Process kworker/u16:4 (pid: 4584, stack limit = 0x00000000c9e45703)
Call trace:
mlx5e_rx_reporter_err_rq_cqe_recover+0x30/0x164 [mlx5_core]
mlx5e_health_report+0x60/0x6c [mlx5_core]
mlx5e_reporter_rq_cqe_err+0x6c/0x90 [mlx5_core]
mlx5e_rq_err_cqe_work+0x20/0x2c [mlx5_core]
process_one_work+0x168/0x3d0
worker_thread+0x58/0x3d0
kthread+0x108/0x134

Fixes: c50de4af1d63 ("net/mlx5e: Generalize tx reporter's functionality")
Signed-off-by: Aya Levin <ayal@mellanox.com>
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Reset RQ doorbell counter before moving RQ state from RST to RDY
Aya Levin [Mon, 9 Dec 2019 12:08:18 +0000 (14:08 +0200)]
net/mlx5e: Reset RQ doorbell counter before moving RQ state from RST to RDY

Initialize RQ doorbell counters to zero prior to moving an RQ from RST
to RDY state. Per HW spec, when RQ is back to RDY state, the descriptor
ID on the completion is reset. The doorbell record must comply.

Fixes: 8276ea1353a4 ("net/mlx5e: Report and recover from CQE with error on RQ")
Signed-off-by: Aya Levin <ayal@mellanox.com>
Reported-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Fix sleep while atomic in mlx5_eswitch_get_vepa
Huy Nguyen [Mon, 3 Feb 2020 22:32:18 +0000 (16:32 -0600)]
net/mlx5: Fix sleep while atomic in mlx5_eswitch_get_vepa

rtnl_bridge_getlink is protected by rcu lock, so mlx5_eswitch_get_vepa
cannot take mutex lock. Two possible issues can happen:
1. User at the same time change vepa mode via RTM_SETLINK command.
2. User at the same time change the switchdev mode via devlink netlink
interface.

Case 1 cannot happen because rtnl executes one message in order.
Case 2 can happen but we do not expect user to change the switchdev mode
when changing vepa. Even if a user does it, so he will read a value
which is no longer valid.

Fixes: 8da202b24913 ("net/mlx5: E-Switch, Add support for VEPA in legacy mode.")
Signed-off-by: Huy Nguyen <huyn@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf
David S. Miller [Tue, 18 Feb 2020 23:44:13 +0000 (15:44 -0800)]
Merge git://git./pub/scm/linux/kernel/git/pablo/nf

Pablo Neira Ayuso says:

====================
Netfilter fixes for net

This batch contains Netfilter fixes for net:

1) Restrict hashlimit size to 1048576, from Cong Wang.

2) Check for offload flags from nf_flow_table_offload_setup(),
   this fixes a crash in case the hardware offload is disabled.
   From Florian Westphal.

3) Three preparation patches to extend the conntrack clash resolution,
   from Florian.

4) Extend clash resolution to deal with DNS packets from the same flow
   racing to set up the NAT configuration.

5) Small documentation fix in pipapo, from Stefano Brivio.

6) Remove misleading unlikely() from pipapo_refill(), also from Stefano.

7) Reduce hashlimit mutex scope, from Cong Wang. This patch is actually
   triggering another problem, still under discussion, another patch to
   fix this will follow up.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonetfilter: nft_set_pipapo: Don't abuse unlikely() in pipapo_refill()
Stefano Brivio [Fri, 14 Feb 2020 17:14:14 +0000 (18:14 +0100)]
netfilter: nft_set_pipapo: Don't abuse unlikely() in pipapo_refill()

I originally used unlikely() in the if (match_only) clause, which
we hit on the mapping table for the last field in a set, to ensure
we avoid branching to the rest of for loop body, which is executed
more frequently.

However, Pablo reports, this is confusing as it gives the impression
that this is not a common case, and it's actually not the intended
usage of unlikely().

I couldn't observe any statistical difference in matching rates on
x864_64 and aarch64 without it, so just drop it.

Reported-by: Pablo Neira Ayuso <pablo@netfilter.org>
Fixes: 3c4287f62044 ("nf_tables: Add set type for arbitrary concatenation of ranges")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
4 years agonetfilter: nft_set_pipapo: Fix mapping table example in comments
Stefano Brivio [Fri, 14 Feb 2020 17:14:13 +0000 (18:14 +0100)]
netfilter: nft_set_pipapo: Fix mapping table example in comments

In both insertion and lookup examples, the two element pointers
of rule mapping tables were swapped. Fix that.

Reported-by: Pablo Neira Ayuso <pablo@netfilter.org>
Fixes: 3c4287f62044 ("nf_tables: Add set type for arbitrary concatenation of ranges")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
4 years agoflow_table.c: Use built-in RCU list checking
Madhuparna Bhowmik [Tue, 18 Feb 2020 19:58:20 +0000 (01:28 +0530)]
flow_table.c: Use built-in RCU list checking

hlist_for_each_entry_rcu() has built-in RCU and lock checking.

Pass cond argument to list_for_each_entry_rcu() to silence
false lockdep warning when CONFIG_PROVE_RCU_LIST is enabled
by default.

Signed-off-by: Madhuparna Bhowmik <madhuparnabhowmik10@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodatapath.c: Use built-in RCU list checking
Madhuparna Bhowmik [Tue, 18 Feb 2020 19:58:02 +0000 (01:28 +0530)]
datapath.c: Use built-in RCU list checking

hlist_for_each_entry_rcu() has built-in RCU and lock checking.

Pass cond argument to list_for_each_entry_rcu() to silence
false lockdep warning when CONFIG_PROVE_RCU_LIST is enabled
by default.

Signed-off-by: Madhuparna Bhowmik <madhuparnabhowmik10@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agovport.c: Use built-in RCU list checking
Madhuparna Bhowmik [Tue, 18 Feb 2020 19:57:42 +0000 (01:27 +0530)]
vport.c: Use built-in RCU list checking

hlist_for_each_entry_rcu() has built-in RCU and lock checking.

Pass cond argument to list_for_each_entry_rcu() to silence
false lockdep warning when CONFIG_PROVE_RCU_LIST is enabled
by default.

Signed-off-by: Madhuparna Bhowmik <madhuparnabhowmik10@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agometer.c: Use built-in RCU list checking
Madhuparna Bhowmik [Tue, 18 Feb 2020 19:54:25 +0000 (01:24 +0530)]
meter.c: Use built-in RCU list checking

hlist_for_each_entry_rcu() has built-in RCU and lock checking.

Pass cond argument to list_for_each_entry_rcu() to silence
false lockdep warning when CONFIG_PROVE_RCU_LIST is enabled
by default.

Signed-off-by: Madhuparna Bhowmik <madhuparnabhowmik10@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonetlabel_domainhash.c: Use built-in RCU list checking
Madhuparna Bhowmik [Tue, 18 Feb 2020 18:41:32 +0000 (00:11 +0530)]
netlabel_domainhash.c: Use built-in RCU list checking

list_for_each_entry_rcu() has built-in RCU and lock checking.

Pass cond argument to list_for_each_entry_rcu() to silence
false lockdep warning when CONFIG_PROVE_RCU_LIST is enabled
by default.

Signed-off-by: Madhuparna Bhowmik <madhuparnabhowmik10@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: netlabel: Use built-in RCU list checking
Madhuparna Bhowmik [Tue, 18 Feb 2020 18:17:18 +0000 (23:47 +0530)]
net: netlabel: Use built-in RCU list checking

list_for_each_entry_rcu() has built-in RCU and lock checking.

Pass cond argument to list_for_each_entry_rcu() to silence
false lockdep warning when CONFIG_PROVE_RCU_LIST is enabled
by default.

Signed-off-by: Madhuparna Bhowmik <madhuparnabhowmik10@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoRevert "net: dev: introduce support for sch BYPASS for lockless qdisc"
Paolo Abeni [Tue, 18 Feb 2020 17:15:44 +0000 (18:15 +0100)]
Revert "net: dev: introduce support for sch BYPASS for lockless qdisc"

This reverts commit ba27b4cdaaa66561aaedb2101876e563738d36fe

Ahmed reported ouf-of-order issues bisected to commit ba27b4cdaaa6
("net: dev: introduce support for sch BYPASS for lockless qdisc").
I can't find any working solution other than a plain revert.

This will introduce some minor performance regressions for
pfifo_fast qdisc. I plan to address them in net-next with more
indirect call wrapper boilerplate for qdiscs.

Reported-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
Fixes: ba27b4cdaaa6 ("net: dev: introduce support for sch BYPASS for lockless qdisc")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: broadcom: Fix a typo ("firsly")
Jonathan Neuschäfer [Tue, 18 Feb 2020 15:47:01 +0000 (16:47 +0100)]
net: phy: broadcom: Fix a typo ("firsly")

Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: fix bogus socket flag values
Florian Westphal [Mon, 17 Feb 2020 15:54:38 +0000 (16:54 +0100)]
mptcp: fix bogus socket flag values

Dan Carpenter reports static checker warnings due to bogus BIT() usage:

net/mptcp/subflow.c:571 subflow_write_space() warn: test_bit() takes a bit number
net/mptcp/subflow.c:694 subflow_state_change() warn: test_bit() takes a bit number
net/mptcp/protocol.c:261 ssk_check_wmem() warn: test_bit() takes a bit number
[..]

This is harmless (we use bits 1 & 2 instead of 0 and 1), but would
break eventually when adding BIT(5) (or 6, depends on size of 'long').

Just use 0 and 1, the values are only passed to test/set/clear_bit
functions.

Fixes: 648ef4b88673 ("mptcp: Implement MPTCP receive path")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoqede: Fix race between rdma destroy workqueue and link change event
Michal Kalderon [Mon, 17 Feb 2020 11:37:18 +0000 (13:37 +0200)]
qede: Fix race between rdma destroy workqueue and link change event

If an event is added while the rdma workqueue is being destroyed
it could lead to several races, list corruption, null pointer
dereference during queue_work or init_queue.
This fixes the race between the two flows which can occur during
shutdown.

A kref object and a completion object are added to the rdma_dev
structure, these are initialized before the workqueue is created.
The refcnt is used to indicate work is being added to the
workqueue and ensures the cleanup flow won't start while we're in
the middle of adding the event.
Once the work is added, the refcnt is decreased and the cleanup flow
is safe to run.

Fixes: cee9fbd8e2e ("qede: Add qedr framework")
Signed-off-by: Ariel Elior <ariel.elior@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agolibbpf: Sanitise internal map names so they are not rejected by the kernel
Toke Høiland-Jørgensen [Mon, 17 Feb 2020 17:17:01 +0000 (18:17 +0100)]
libbpf: Sanitise internal map names so they are not rejected by the kernel

The kernel only accepts map names with alphanumeric characters, underscores
and periods in their name. However, the auto-generated internal map names
used by libbpf takes their prefix from the user-supplied BPF object name,
which has no such restriction. This can lead to "Invalid argument" errors
when trying to load a BPF program using global variables.

Fix this by sanitising the map names, replacing any non-allowed characters
with underscores.

Fixes: d859900c4c56 ("bpf, libbpf: support global data/bss/rodata sections")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200217171701.215215-1-toke@redhat.com
4 years agobpf, uapi: Remove text about bpf_redirect_map() giving higher performance
Toke Høiland-Jørgensen [Tue, 18 Feb 2020 13:03:34 +0000 (14:03 +0100)]
bpf, uapi: Remove text about bpf_redirect_map() giving higher performance

The performance of bpf_redirect() is now roughly the same as that of
bpf_redirect_map(). However, David Ahern pointed out that the header file
has not been updated to reflect this, and still says that a significant
performance increase is possible when using bpf_redirect_map(). Remove this
text from the bpf_redirect_map() description, and reword the description in
bpf_redirect() slightly. Also fix the 'Return' section of the
bpf_redirect_map() documentation.

Fixes: 1d233886dd90 ("xdp: Use bulking for non-map XDP_REDIRECT and consolidate code paths")
Reported-by: David Ahern <dsahern@gmail.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20200218130334.29889-1-toke@redhat.com
4 years agoselftests: forwarding: vxlan_bridge_1d: use more proper tos value
Hangbin Liu [Mon, 17 Feb 2020 03:43:15 +0000 (11:43 +0800)]
selftests: forwarding: vxlan_bridge_1d: use more proper tos value

0x11 and 0x12 set the ECN bits based on RFC2474, it would be better to avoid
that. 0x14 and 0x18 would be better and works as well.

Reported-by: Petr Machata <petrm@mellanox.com>
Fixes: 4e867c9a50ff ("selftests: forwarding: vxlan_bridge_1d: fix tos value")
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: cnic: fix spelling mistake "reserverd" -> "reserved"
Alexandre Belloni [Fri, 14 Feb 2020 14:32:24 +0000 (15:32 +0100)]
net: cnic: fix spelling mistake "reserverd" -> "reserved"

The reserved member should be named reserved3.

Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosctp: move the format error check out of __sctp_sf_do_9_1_abort
Xin Long [Tue, 18 Feb 2020 04:07:53 +0000 (12:07 +0800)]
sctp: move the format error check out of __sctp_sf_do_9_1_abort

When T2 timer is to be stopped, the asoc should also be deleted,
otherwise, there will be no chance to call sctp_association_free
and the asoc could last in memory forever.

However, in sctp_sf_shutdown_sent_abort(), after adding the cmd
SCTP_CMD_TIMER_STOP for T2 timer, it may return error due to the
format error from __sctp_sf_do_9_1_abort() and miss adding
SCTP_CMD_ASSOC_FAILED where the asoc will be deleted.

This patch is to fix it by moving the format error check out of
__sctp_sf_do_9_1_abort(), and do it before adding the cmd
SCTP_CMD_TIMER_STOP for T2 timer.

Thanks Hangbin for reporting this issue by the fuzz testing.

v1->v2:
  - improve the comment in the code as Marcelo's suggestion.

Fixes: 96ca468b86b0 ("sctp: check invalid value of length parameter in error cause")
Reported-by: Hangbin Liu <liuhangbin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: sched: correct flower port blocking
Jason Baron [Mon, 17 Feb 2020 20:38:09 +0000 (15:38 -0500)]
net: sched: correct flower port blocking

tc flower rules that are based on src or dst port blocking are sometimes
ineffective due to uninitialized stack data. __skb_flow_dissect() extracts
ports from the skb for tc flower to match against. However, the port
dissection is not done when when the FLOW_DIS_IS_FRAGMENT bit is set in
key_control->flags. All callers of __skb_flow_dissect(), zero-out the
key_control field except for fl_classify() as used by the flower
classifier. Thus, the FLOW_DIS_IS_FRAGMENT may be set on entry to
__skb_flow_dissect(), since key_control is allocated on the stack
and may not be initialized.

Since key_basic and key_control are present for all flow keys, let's
make sure they are initialized.

Fixes: 62230715fd24 ("flow_dissector: do not dissect l4 ports for fragments")
Co-developed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jason Baron <jbaron@akamai.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mscc: fix in frame extraction
Horatiu Vultur [Mon, 17 Feb 2020 08:31:33 +0000 (09:31 +0100)]
net: mscc: fix in frame extraction

Each extracted frame on Ocelot has an IFH. The frame and IFH are extracted
by reading chuncks of 4 bytes from a register.

In case the IFH and frames were read corretly it would try to read the next
frame. In case there are no more frames in the queue, it checks if there
were any previous errors and in that case clear the queue. But this check
will always succeed also when there are no errors. Because when extracting
the IFH the error is checked against 4(number of bytes read) and then the
error is set only if the extraction of the frame failed. So in a happy case
where there are no errors the err variable is still 4. So it could be
a case where after the check that there are no more frames in the queue, a
frame will arrive in the queue but because the error is not reseted, it
would try to flush the queue. So the frame will be lost.

The fix consist in resetting the error after reading the IFH.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Acked-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobpf, offload: Replace bitwise AND by logical AND in bpf_prog_offload_info_fill
Johannes Krude [Wed, 12 Feb 2020 19:32:27 +0000 (20:32 +0100)]
bpf, offload: Replace bitwise AND by logical AND in bpf_prog_offload_info_fill

This if guards whether user-space wants a copy of the offload-jited
bytecode and whether this bytecode exists. By erroneously doing a bitwise
AND instead of a logical AND on user- and kernel-space buffer-size can lead
to no data being copied to user-space especially when user-space size is a
power of two and bigger then the kernel-space buffer.

Fixes: fcfb126defda ("bpf: add new jited info fields in bpf_dev_offload and bpf_prog_info")
Signed-off-by: Johannes Krude <johannes@krude.de>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/bpf/20200212193227.GA3769@phlox.h.transitiv.net
4 years agonetfilter: conntrack: allow insertion of clashing entries
Florian Westphal [Mon, 3 Feb 2020 16:37:07 +0000 (17:37 +0100)]
netfilter: conntrack: allow insertion of clashing entries

This patch further relaxes the need to drop an skb due to a clash with
an existing conntrack entry.

Current clash resolution handles the case where the clash occurs between
two identical entries (distinct nf_conn objects with same tuples), i.e.:

                    Original                        Reply
existing: 10.2.3.4:42 -> 10.8.8.8:53      10.2.3.4:42 <- 10.0.0.6:5353
clashing: 10.2.3.4:42 -> 10.8.8.8:53      10.2.3.4:42 <- 10.0.0.6:5353

... existing handling will discard the unconfirmed clashing entry and
makes skb->_nfct point to the existing one.  The skb can then be
processed normally just as if the clash would not have existed in the
first place.

For other clashes, the skb needs to be dropped.
This frequently happens with DNS resolvers that send A and AAAA queries
back-to-back when NAT rules are present that cause packets to get
different DNAT transformations applied, for example:

-m statistics --mode random ... -j DNAT --dnat-to 10.0.0.6:5353
-m statistics --mode random ... -j DNAT --dnat-to 10.0.0.7:5353

In this case the A or AAAA query is dropped which incurs a costly
delay during name resolution.

This patch also allows this collision type:
                       Original                   Reply
existing: 10.2.3.4:42 -> 10.8.8.8:53      10.2.3.4:42 <- 10.0.0.6:5353
clashing: 10.2.3.4:42 -> 10.8.8.8:53      10.2.3.4:42 <- 10.0.0.7:5353

In this case, clash is in original direction -- the reply direction
is still unique.

The change makes it so that when the 2nd colliding packet is received,
the clashing conntrack is tagged with new IPS_NAT_CLASH_BIT, gets a fixed
1 second timeout and is inserted in the reply direction only.

The entry is hidden from 'conntrack -L', it will time out quickly
and it can be early dropped because it will never progress to the
ASSURED state.

To avoid special-casing the delete code path to special case
the ORIGINAL hlist_nulls node, a new helper, "hlist_nulls_add_fake", is
added so hlist_nulls_del() will work.

Example:

      CPU A:                               CPU B:
1.  10.2.3.4:42 -> 10.8.8.8:53 (A)
2.                                         10.2.3.4:42 -> 10.8.8.8:53 (AAAA)
3.  Apply DNAT, reply changed to 10.0.0.6
4.                                         10.2.3.4:42 -> 10.8.8.8:53 (AAAA)
5.                                         Apply DNAT, reply changed to 10.0.0.7
6. confirm/commit to conntrack table, no collisions
7.                                         commit clashing entry

Reply comes in:

10.2.3.4:42 <- 10.0.0.6:5353 (A)
 -> Finds a conntrack, DNAT is reversed & packet forwarded to 10.2.3.4:42
10.2.3.4:42 <- 10.0.0.7:5353 (AAAA)
 -> Finds a conntrack, DNAT is reversed & packet forwarded to 10.2.3.4:42
    The conntrack entry is deleted from table, as it has the NAT_CLASH
    bit set.

In case of a retransmit from ORIGINAL dir, all further packets will get
the DNAT transformation to 10.0.0.6.

I tried to come up with other solutions but they all have worse
problems.

Alternatives considered were:
1.  Confirm ct entries at allocation time, not in postrouting.
 a. will cause uneccesarry work when the skb that creates the
    conntrack is dropped by ruleset.
 b. in case nat is applied, ct entry would need to be moved in
    the table, which requires another spinlock pair to be taken.
 c. breaks the 'unconfirmed entry is private to cpu' assumption:
    we would need to guard all nfct->ext allocation requests with
    ct->lock spinlock.

2. Make the unconfirmed list a hash table instead of a pcpu list.
   Shares drawback c) of the first alternative.

3. Document this is expected and force users to rearrange their
   ruleset (e.g. by using "-m cluster" instead of "-m statistics").
   nft has the 'jhash' expression which can be used instead of 'numgen'.

   Major drawback: doesn't fix what I consider a bug, not very realistic
   and I believe its reasonable to have the existing rulesets to 'just
   work'.

4. Document this is expected and force users to steer problematic
   packets to the same CPU -- this would serialize the "allocate new
   conntrack entry/nat table evaluation/perform nat/confirm entry", so
   no race can occur.  Similar drawback to 3.

Another advantage of this patch compared to 1) and 2) is that there are
no changes to the hot path; things are handled in the udp tracker and
the clash resolution path.

Cc: rcu@vger.kernel.org
Cc: "Paul E. McKenney" <paulmck@kernel.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Jozsef Kadlecsik <kadlec@netfilter.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
4 years agonet: ethernet: dm9000: Handle -EPROBE_DEFER in dm9000_parse_dt()
Paul Cercueil [Sun, 16 Feb 2020 19:39:43 +0000 (16:39 -0300)]
net: ethernet: dm9000: Handle -EPROBE_DEFER in dm9000_parse_dt()

The call to of_get_mac_address() can return -EPROBE_DEFER, for instance
when the MAC address is read from a NVMEM driver that did not probe yet.

Cc: H. Nikolaus Schaller <hns@goldelico.com>
Cc: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Paul Cercueil <paul@crapouillou.net>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoskbuff.h: fix all kernel-doc warnings
Randy Dunlap [Sat, 15 Feb 2020 23:34:07 +0000 (15:34 -0800)]
skbuff.h: fix all kernel-doc warnings

Fix all kernel-doc warnings in <linux/skbuff.h>.
Fixes these warnings:

../include/linux/skbuff.h:890: warning: Function parameter or member 'list' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'dev_scratch' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'ip_defrag_offset' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'skb_mstamp_ns' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member '__cloned_offset' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'head_frag' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member '__pkt_type_offset' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'encapsulation' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'encap_hdr_csum' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'csum_valid' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member '__pkt_vlan_present_offset' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'vlan_present' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'csum_complete_sw' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'csum_level' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'inner_protocol_type' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'remcsum_offload' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'sender_cpu' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'reserved_tailroom' not described in 'sk_buff'
../include/linux/skbuff.h:890: warning: Function parameter or member 'inner_ipproto' not described in 'sk_buff'

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoskbuff: remove stale bit mask comments
Randy Dunlap [Sat, 15 Feb 2020 21:41:12 +0000 (13:41 -0800)]
skbuff: remove stale bit mask comments

Remove stale comments since this flag is no longer a bit mask
but is a bit field.

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/sock.h: fix all kernel-doc warnings
Randy Dunlap [Sat, 15 Feb 2020 19:42:37 +0000 (11:42 -0800)]
net/sock.h: fix all kernel-doc warnings

Fix all kernel-doc warnings for <net/sock.h>.
Fixes these warnings:

../include/net/sock.h:232: warning: Function parameter or member 'skc_addrpair' not described in 'sock_common'
../include/net/sock.h:232: warning: Function parameter or member 'skc_portpair' not described in 'sock_common'
../include/net/sock.h:232: warning: Function parameter or member 'skc_ipv6only' not described in 'sock_common'
../include/net/sock.h:232: warning: Function parameter or member 'skc_net_refcnt' not described in 'sock_common'
../include/net/sock.h:232: warning: Function parameter or member 'skc_v6_daddr' not described in 'sock_common'
../include/net/sock.h:232: warning: Function parameter or member 'skc_v6_rcv_saddr' not described in 'sock_common'
../include/net/sock.h:232: warning: Function parameter or member 'skc_cookie' not described in 'sock_common'
../include/net/sock.h:232: warning: Function parameter or member 'skc_listener' not described in 'sock_common'
../include/net/sock.h:232: warning: Function parameter or member 'skc_tw_dr' not described in 'sock_common'
../include/net/sock.h:232: warning: Function parameter or member 'skc_rcv_wnd' not described in 'sock_common'
../include/net/sock.h:232: warning: Function parameter or member 'skc_tw_rcv_nxt' not described in 'sock_common'

../include/net/sock.h:498: warning: Function parameter or member 'sk_rx_skb_cache' not described in 'sock'
../include/net/sock.h:498: warning: Function parameter or member 'sk_wq_raw' not described in 'sock'
../include/net/sock.h:498: warning: Function parameter or member 'tcp_rtx_queue' not described in 'sock'
../include/net/sock.h:498: warning: Function parameter or member 'sk_tx_skb_cache' not described in 'sock'
../include/net/sock.h:498: warning: Function parameter or member 'sk_route_forced_caps' not described in 'sock'
../include/net/sock.h:498: warning: Function parameter or member 'sk_txtime_report_errors' not described in 'sock'
../include/net/sock.h:498: warning: Function parameter or member 'sk_validate_xmit_skb' not described in 'sock'
../include/net/sock.h:498: warning: Function parameter or member 'sk_bpf_storage' not described in 'sock'

../include/net/sock.h:2024: warning: No description found for return value of 'sk_wmem_alloc_get'
../include/net/sock.h:2035: warning: No description found for return value of 'sk_rmem_alloc_get'
../include/net/sock.h:2046: warning: No description found for return value of 'sk_has_allocations'
../include/net/sock.h:2082: warning: No description found for return value of 'skwq_has_sleeper'
../include/net/sock.h:2244: warning: No description found for return value of 'sk_page_frag'
../include/net/sock.h:2444: warning: Function parameter or member 'tcp_rx_skb_cache_key' not described in 'DECLARE_STATIC_KEY_FALSE'
../include/net/sock.h:2444: warning: Excess function parameter 'sk' description in 'DECLARE_STATIC_KEY_FALSE'
../include/net/sock.h:2444: warning: Excess function parameter 'skb' description in 'DECLARE_STATIC_KEY_FALSE'

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ks8851-ml: Fix 16-bit IO operation
Marek Vasut [Sat, 15 Feb 2020 16:54:19 +0000 (17:54 +0100)]
net: ks8851-ml: Fix 16-bit IO operation

The Micrel KSZ8851-16MLLI datasheet DS00002357B page 12 states that
BE[3:0] signals are active high. This contradicts the measurements
of the behavior of the actual chip, where these signals behave as
active low. For example, to read the CIDER register, the bus must
expose 0xc0c0 during the address phase, which means BE[3:0]=4'b1100.

Signed-off-by: Marek Vasut <marex@denx.de>
Cc: David S. Miller <davem@davemloft.net>
Cc: Lukas Wunner <lukas@wunner.de>
Cc: Petr Stetiar <ynezz@true.cz>
Cc: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ks8851-ml: Fix 16-bit data access
Marek Vasut [Sat, 15 Feb 2020 16:54:18 +0000 (17:54 +0100)]
net: ks8851-ml: Fix 16-bit data access

The packet data written to and read from Micrel KSZ8851-16MLLI must be
byte-swapped in 16-bit mode, add this byte-swapping.

Signed-off-by: Marek Vasut <marex@denx.de>
Cc: David S. Miller <davem@davemloft.net>
Cc: Lukas Wunner <lukas@wunner.de>
Cc: Petr Stetiar <ynezz@true.cz>
Cc: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ks8851-ml: Remove 8-bit bus accessors
Marek Vasut [Sat, 15 Feb 2020 16:54:17 +0000 (17:54 +0100)]
net: ks8851-ml: Remove 8-bit bus accessors

This driver is mixing 8-bit and 16-bit bus accessors for reasons unknown,
however the speculation is that this was some sort of attempt to support
the 8-bit bus mode.

As per the KS8851-16MLL documentation, all two registers accessed via the
8-bit accessors are internally 16-bit registers, so reading them using
16-bit accessors is fine. The KS_CCR read can be converted to 16-bit read
outright, as it is already a concatenation of two 8-bit reads of that
register. The KS_RXQCR accesses are 8-bit only, however writing the top
8 bits of the register is OK as well, since the driver caches the entire
16-bit register value anyway.

Finally, the driver is not used by any hardware in the kernel right now.
The only hardware available to me is one with 16-bit bus, so I have no
way to test the 8-bit bus mode, however it is unlikely this ever really
worked anyway. If the 8-bit bus mode is ever required, it can be easily
added by adjusting the 16-bit accessors to do 2 consecutive accesses,
which is how this should have been done from the beginning.

Signed-off-by: Marek Vasut <marex@denx.de>
Cc: David S. Miller <davem@davemloft.net>
Cc: Lukas Wunner <lukas@wunner.de>
Cc: Petr Stetiar <ynezz@true.cz>
Cc: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: select CRYPTO
Matthieu Baerts [Sat, 15 Feb 2020 14:45:56 +0000 (15:45 +0100)]
mptcp: select CRYPTO

Without this modification and if CRYPTO is not selected, we have this
warning:

  WARNING: unmet direct dependencies detected for CRYPTO_LIB_SHA256
    Depends on [n]: CRYPTO [=n]
    Selected by [y]:
    - MPTCP [=y] && NET [=y] && INET [=y]

MPTCP selects CRYPTO_LIB_SHA256 which seems to depend on CRYPTO. CRYPTO
is now selected to avoid this issue.

Even though the config system prints that warning, it looks like
sha256.c is compiled and linked even without CONFIG_CRYPTO. Since MPTCP
will end up needing CONFIG_CRYPTO anyway in future commits -- currently
in preparation for net-next -- we propose to add it now to fix the
warning.

The dependency in the config system comes from the fact that
CRYPTO_LIB_SHA256 is defined in "lib/crypto/Kconfig" which is sourced
from "crypto/Kconfig" only if CRYPTO is selected.

Fixes: 65492c5a6ab5 (mptcp: move from sha1 (v0) to sha256 (v1))
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'bonding-fix-bonding-interface-bugs'
David S. Miller [Mon, 17 Feb 2020 03:32:11 +0000 (19:32 -0800)]
Merge branch 'bonding-fix-bonding-interface-bugs'

Taehee Yoo says:

====================
bonding: fix bonding interface bugs

This patchset fixes lockdep problem in bonding interface

1. The first patch is to add missing netdev_update_lockdep_key().
After bond_release(), netdev_update_lockdep_key() should be called.
But both ioctl path and attribute path don't call
netdev_update_lockdep_key().
This patch adds missing netdev_update_lockdep_key().

2. The second patch is to export netdev_next_lower_dev_rcu symbol.
netdev_next_lower_dev_rcu() is useful to implement the function,
which is to walk their all lower interfaces.
This patch is actually a preparing patch for the third patch.

3. The last patch is to fix lockdep waring in bond_get_stats().
The stats_lock uses a dynamic lockdep key.
So, after "nomaster" operation, updating the dynamic lockdep key
routine is needed. but it doesn't
So, lockdep warning occurs.

Change log:
v1 -> v2:
 - Update headline from "fix bonding interface bugs"
   to "bonding: fix bonding interface bugs"
 - Drop a patch("bonding: do not collect slave's stats")
 - Add new patches
   - ("net: export netdev_next_lower_dev_rcu()")
   - ("bonding: fix lockdep warning in bond_get_stats()")
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobonding: fix lockdep warning in bond_get_stats()
Taehee Yoo [Sat, 15 Feb 2020 10:50:40 +0000 (10:50 +0000)]
bonding: fix lockdep warning in bond_get_stats()

In the "struct bonding", there is stats_lock.
This lock protects "bond_stats" in the "struct bonding".
bond_stats is updated in the bond_get_stats() and this function would be
executed concurrently. So, the lock is needed.

Bonding interfaces would be nested.
So, either stats_lock should use dynamic lockdep class key or stats_lock
should be used by spin_lock_nested(). In the current code, stats_lock is
using a dynamic lockdep class key.
But there is no updating stats_lock_key routine So, lockdep warning
will occur.

Test commands:
    ip link add bond0 type bond
    ip link add bond1 type bond
    ip link set bond0 master bond1
    ip link set bond0 nomaster
    ip link set bond1 master bond0

Splat looks like:
[   38.420603][  T957] 5.5.0+ #394 Not tainted
[   38.421074][  T957] ------------------------------------------------------
[   38.421837][  T957] ip/957 is trying to acquire lock:
[   38.422399][  T957] ffff888063262cd8 (&bond->stats_lock_key#2){+.+.}, at: bond_get_stats+0x90/0x4d0 [bonding]
[   38.423528][  T957]
[   38.423528][  T957] but task is already holding lock:
[   38.424526][  T957] ffff888065fd2cd8 (&bond->stats_lock_key){+.+.}, at: bond_get_stats+0x90/0x4d0 [bonding]
[   38.426075][  T957]
[   38.426075][  T957] which lock already depends on the new lock.
[   38.426075][  T957]
[   38.428536][  T957]
[   38.428536][  T957] the existing dependency chain (in reverse order) is:
[   38.429475][  T957]
[   38.429475][  T957] -> #1 (&bond->stats_lock_key){+.+.}:
[   38.430273][  T957]        _raw_spin_lock+0x30/0x70
[   38.430812][  T957]        bond_get_stats+0x90/0x4d0 [bonding]
[   38.431451][  T957]        dev_get_stats+0x1ec/0x270
[   38.432088][  T957]        bond_get_stats+0x1a5/0x4d0 [bonding]
[   38.432767][  T957]        dev_get_stats+0x1ec/0x270
[   38.433322][  T957]        rtnl_fill_stats+0x44/0xbe0
[   38.433866][  T957]        rtnl_fill_ifinfo+0xeb2/0x3720
[   38.434474][  T957]        rtmsg_ifinfo_build_skb+0xca/0x170
[   38.435081][  T957]        rtmsg_ifinfo_event.part.33+0x1b/0xb0
[   38.436848][  T957]        rtnetlink_event+0xcd/0x120
[   38.437455][  T957]        notifier_call_chain+0x90/0x160
[   38.438067][  T957]        netdev_change_features+0x74/0xa0
[   38.438708][  T957]        bond_compute_features.isra.45+0x4e6/0x6f0 [bonding]
[   38.439522][  T957]        bond_enslave+0x3639/0x47b0 [bonding]
[   38.440225][  T957]        do_setlink+0xaab/0x2ef0
[   38.440786][  T957]        __rtnl_newlink+0x9c5/0x1270
[   38.441463][  T957]        rtnl_newlink+0x65/0x90
[   38.442075][  T957]        rtnetlink_rcv_msg+0x4a8/0x890
[   38.442774][  T957]        netlink_rcv_skb+0x121/0x350
[   38.443451][  T957]        netlink_unicast+0x42e/0x610
[   38.444282][  T957]        netlink_sendmsg+0x65a/0xb90
[   38.444992][  T957]        ____sys_sendmsg+0x5ce/0x7a0
[   38.445679][  T957]        ___sys_sendmsg+0x10f/0x1b0
[   38.446365][  T957]        __sys_sendmsg+0xc6/0x150
[   38.447007][  T957]        do_syscall_64+0x99/0x4f0
[   38.447668][  T957]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
[   38.448538][  T957]
[   38.448538][  T957] -> #0 (&bond->stats_lock_key#2){+.+.}:
[   38.449554][  T957]        __lock_acquire+0x2d8d/0x3de0
[   38.450148][  T957]        lock_acquire+0x164/0x3b0
[   38.450711][  T957]        _raw_spin_lock+0x30/0x70
[   38.451292][  T957]        bond_get_stats+0x90/0x4d0 [bonding]
[   38.451950][  T957]        dev_get_stats+0x1ec/0x270
[   38.452425][  T957]        bond_get_stats+0x1a5/0x4d0 [bonding]
[   38.453362][  T957]        dev_get_stats+0x1ec/0x270
[   38.453825][  T957]        rtnl_fill_stats+0x44/0xbe0
[   38.454390][  T957]        rtnl_fill_ifinfo+0xeb2/0x3720
[   38.456257][  T957]        rtmsg_ifinfo_build_skb+0xca/0x170
[   38.456998][  T957]        rtmsg_ifinfo_event.part.33+0x1b/0xb0
[   38.459351][  T957]        rtnetlink_event+0xcd/0x120
[   38.460086][  T957]        notifier_call_chain+0x90/0x160
[   38.460829][  T957]        netdev_change_features+0x74/0xa0
[   38.461752][  T957]        bond_compute_features.isra.45+0x4e6/0x6f0 [bonding]
[   38.462705][  T957]        bond_enslave+0x3639/0x47b0 [bonding]
[   38.463476][  T957]        do_setlink+0xaab/0x2ef0
[   38.464141][  T957]        __rtnl_newlink+0x9c5/0x1270
[   38.464897][  T957]        rtnl_newlink+0x65/0x90
[   38.465522][  T957]        rtnetlink_rcv_msg+0x4a8/0x890
[   38.466215][  T957]        netlink_rcv_skb+0x121/0x350
[   38.466895][  T957]        netlink_unicast+0x42e/0x610
[   38.467583][  T957]        netlink_sendmsg+0x65a/0xb90
[   38.468285][  T957]        ____sys_sendmsg+0x5ce/0x7a0
[   38.469202][  T957]        ___sys_sendmsg+0x10f/0x1b0
[   38.469884][  T957]        __sys_sendmsg+0xc6/0x150
[   38.470587][  T957]        do_syscall_64+0x99/0x4f0
[   38.471245][  T957]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
[   38.472093][  T957]
[   38.472093][  T957] other info that might help us debug this:
[   38.472093][  T957]
[   38.473438][  T957]  Possible unsafe locking scenario:
[   38.473438][  T957]
[   38.474898][  T957]        CPU0                    CPU1
[   38.476234][  T957]        ----                    ----
[   38.480171][  T957]   lock(&bond->stats_lock_key);
[   38.480808][  T957]                                lock(&bond->stats_lock_key#2);
[   38.481791][  T957]                                lock(&bond->stats_lock_key);
[   38.482754][  T957]   lock(&bond->stats_lock_key#2);
[   38.483416][  T957]
[   38.483416][  T957]  *** DEADLOCK ***
[   38.483416][  T957]
[   38.484505][  T957] 3 locks held by ip/957:
[   38.485048][  T957]  #0: ffffffffbccf6230 (rtnl_mutex){+.+.}, at: rtnetlink_rcv_msg+0x457/0x890
[   38.486198][  T957]  #1: ffff888065fd2cd8 (&bond->stats_lock_key){+.+.}, at: bond_get_stats+0x90/0x4d0 [bonding]
[   38.487625][  T957]  #2: ffffffffbc9254c0 (rcu_read_lock){....}, at: bond_get_stats+0x5/0x4d0 [bonding]
[   38.488897][  T957]
[   38.488897][  T957] stack backtrace:
[   38.489646][  T957] CPU: 1 PID: 957 Comm: ip Not tainted 5.5.0+ #394
[   38.490497][  T957] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[   38.492810][  T957] Call Trace:
[   38.493219][  T957]  dump_stack+0x96/0xdb
[   38.493709][  T957]  check_noncircular+0x371/0x450
[   38.494344][  T957]  ? lookup_address+0x60/0x60
[   38.494923][  T957]  ? print_circular_bug.isra.35+0x310/0x310
[   38.495699][  T957]  ? hlock_class+0x130/0x130
[   38.496334][  T957]  ? __lock_acquire+0x2d8d/0x3de0
[   38.496979][  T957]  __lock_acquire+0x2d8d/0x3de0
[   38.497607][  T957]  ? register_lock_class+0x14d0/0x14d0
[   38.498333][  T957]  ? check_chain_key+0x236/0x5d0
[   38.499003][  T957]  lock_acquire+0x164/0x3b0
[   38.499800][  T957]  ? bond_get_stats+0x90/0x4d0 [bonding]
[   38.500706][  T957]  _raw_spin_lock+0x30/0x70
[   38.501435][  T957]  ? bond_get_stats+0x90/0x4d0 [bonding]
[   38.502311][  T957]  bond_get_stats+0x90/0x4d0 [bonding]
[ ... ]

But, there is another problem.
The dynamic lockdep class key is protected by RTNL, but bond_get_stats()
would be called outside of RTNL.
So, it would use an invalid dynamic lockdep class key.

In order to fix this issue, stats_lock uses spin_lock_nested() instead of
a dynamic lockdep key.
The bond_get_stats() calls bond_get_lowest_level_rcu() to get the correct
nest level value, which will be used by spin_lock_nested().
The "dev->lower_level" indicates lower nest level value, but this value
is invalid outside of RTNL.
So, bond_get_lowest_level_rcu() returns valid lower nest level value in
the RCU critical section.
bond_get_lowest_level_rcu() will be work only when LOCKDEP is enabled.

Fixes: 089bca2caed0 ("bonding: use dynamic lockdep key instead of subclass")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: export netdev_next_lower_dev_rcu()
Taehee Yoo [Sat, 15 Feb 2020 10:50:21 +0000 (10:50 +0000)]
net: export netdev_next_lower_dev_rcu()

netdev_next_lower_dev_rcu() will be used to implement a function,
which is to walk all lower interfaces.
There are already functions that they walk their lower interface.
(netdev_walk_all_lower_dev_rcu, netdev_walk_all_lower_dev()).
But, there would be cases that couldn't be covered by given
netdev_walk_all_lower_dev_{rcu}() function.
So, some modules would want to implement own function,
which is to walk all lower interfaces.

In the next patch, netdev_next_lower_dev_rcu() will be used.
In addition, this patch removes two unused prototypes in netdevice.h.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobonding: add missing netdev_update_lockdep_key()
Taehee Yoo [Sat, 15 Feb 2020 10:50:08 +0000 (10:50 +0000)]
bonding: add missing netdev_update_lockdep_key()

After bond_release(), netdev_update_lockdep_key() should be called.
But both ioctl path and attribute path don't call
netdev_update_lockdep_key().
This patch adds missing netdev_update_lockdep_key().

Test commands:
    ip link add bond0 type bond
    ip link add bond1 type bond
    ifenslave bond0 bond1
    ifenslave -d bond0 bond1
    ifenslave bond1 bond0

Splat looks like:
[   29.501182][ T1046] WARNING: possible circular locking dependency detected
[   29.501945][ T1039] hardirqs last disabled at (1962): [<ffffffffac6c807f>] handle_mm_fault+0x13f/0x700
[   29.503442][ T1046] 5.5.0+ #322 Not tainted
[   29.503447][ T1046] ------------------------------------------------------
[   29.504277][ T1039] softirqs last  enabled at (1180): [<ffffffffade00678>] __do_softirq+0x678/0x981
[   29.505443][ T1046] ifenslave/1046 is trying to acquire lock:
[   29.505886][ T1039] softirqs last disabled at (1169): [<ffffffffac19c18a>] irq_exit+0x17a/0x1a0
[   29.509997][ T1046] ffff88805d5da280 (&dev->addr_list_lock_key#3){+...}, at: dev_mc_sync_multiple+0x95/0x120
[   29.511243][ T1046]
[   29.511243][ T1046] but task is already holding lock:
[   29.512192][ T1046] ffff8880460f2280 (&dev->addr_list_lock_key#4){+...}, at: bond_enslave+0x4482/0x47b0 [bonding]
[   29.514124][ T1046]
[   29.514124][ T1046] which lock already depends on the new lock.
[   29.514124][ T1046]
[   29.517297][ T1046]
[   29.517297][ T1046] the existing dependency chain (in reverse order) is:
[   29.518231][ T1046]
[   29.518231][ T1046] -> #1 (&dev->addr_list_lock_key#4){+...}:
[   29.519076][ T1046]        _raw_spin_lock+0x30/0x70
[   29.519588][ T1046]        dev_mc_sync_multiple+0x95/0x120
[   29.520208][ T1046]        bond_enslave+0x448d/0x47b0 [bonding]
[   29.520862][ T1046]        bond_option_slaves_set+0x1a3/0x370 [bonding]
[   29.521640][ T1046]        __bond_opt_set+0x1ff/0xbb0 [bonding]
[   29.522438][ T1046]        __bond_opt_set_notify+0x2b/0xf0 [bonding]
[   29.523251][ T1046]        bond_opt_tryset_rtnl+0x92/0xf0 [bonding]
[   29.524082][ T1046]        bonding_sysfs_store_option+0x8a/0xf0 [bonding]
[   29.524959][ T1046]        kernfs_fop_write+0x276/0x410
[   29.525620][ T1046]        vfs_write+0x197/0x4a0
[   29.526218][ T1046]        ksys_write+0x141/0x1d0
[   29.526818][ T1046]        do_syscall_64+0x99/0x4f0
[   29.527430][ T1046]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
[   29.528265][ T1046]
[   29.528265][ T1046] -> #0 (&dev->addr_list_lock_key#3){+...}:
[   29.529272][ T1046]        __lock_acquire+0x2d8d/0x3de0
[   29.529935][ T1046]        lock_acquire+0x164/0x3b0
[   29.530638][ T1046]        _raw_spin_lock+0x30/0x70
[   29.531187][ T1046]        dev_mc_sync_multiple+0x95/0x120
[   29.531790][ T1046]        bond_enslave+0x448d/0x47b0 [bonding]
[   29.532451][ T1046]        bond_option_slaves_set+0x1a3/0x370 [bonding]
[   29.533163][ T1046]        __bond_opt_set+0x1ff/0xbb0 [bonding]
[   29.533789][ T1046]        __bond_opt_set_notify+0x2b/0xf0 [bonding]
[   29.534595][ T1046]        bond_opt_tryset_rtnl+0x92/0xf0 [bonding]
[   29.535500][ T1046]        bonding_sysfs_store_option+0x8a/0xf0 [bonding]
[   29.536379][ T1046]        kernfs_fop_write+0x276/0x410
[   29.537057][ T1046]        vfs_write+0x197/0x4a0
[   29.537640][ T1046]        ksys_write+0x141/0x1d0
[   29.538251][ T1046]        do_syscall_64+0x99/0x4f0
[   29.538870][ T1046]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
[   29.539659][ T1046]
[   29.539659][ T1046] other info that might help us debug this:
[   29.539659][ T1046]
[   29.540953][ T1046]  Possible unsafe locking scenario:
[   29.540953][ T1046]
[   29.541883][ T1046]        CPU0                    CPU1
[   29.542540][ T1046]        ----                    ----
[   29.543209][ T1046]   lock(&dev->addr_list_lock_key#4);
[   29.543880][ T1046]                                lock(&dev->addr_list_lock_key#3);
[   29.544873][ T1046]                                lock(&dev->addr_list_lock_key#4);
[   29.545863][ T1046]   lock(&dev->addr_list_lock_key#3);
[   29.546525][ T1046]
[   29.546525][ T1046]  *** DEADLOCK ***
[   29.546525][ T1046]
[   29.547542][ T1046] 5 locks held by ifenslave/1046:
[   29.548196][ T1046]  #0: ffff88806044c478 (sb_writers#5){.+.+}, at: vfs_write+0x3bb/0x4a0
[   29.549248][ T1046]  #1: ffff88805af00890 (&of->mutex){+.+.}, at: kernfs_fop_write+0x1cf/0x410
[   29.550343][ T1046]  #2: ffff88805b8b54b0 (kn->count#157){.+.+}, at: kernfs_fop_write+0x1f2/0x410
[   29.551575][ T1046]  #3: ffffffffaecf4cf0 (rtnl_mutex){+.+.}, at: bond_opt_tryset_rtnl+0x5f/0xf0 [bonding]
[   29.552819][ T1046]  #4: ffff8880460f2280 (&dev->addr_list_lock_key#4){+...}, at: bond_enslave+0x4482/0x47b0 [bonding]
[   29.554175][ T1046]
[   29.554175][ T1046] stack backtrace:
[   29.554907][ T1046] CPU: 0 PID: 1046 Comm: ifenslave Not tainted 5.5.0+ #322
[   29.555854][ T1046] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[   29.557064][ T1046] Call Trace:
[   29.557504][ T1046]  dump_stack+0x96/0xdb
[   29.558054][ T1046]  check_noncircular+0x371/0x450
[   29.558723][ T1046]  ? print_circular_bug.isra.35+0x310/0x310
[   29.559486][ T1046]  ? hlock_class+0x130/0x130
[   29.560100][ T1046]  ? __lock_acquire+0x2d8d/0x3de0
[   29.560761][ T1046]  __lock_acquire+0x2d8d/0x3de0
[   29.561366][ T1046]  ? register_lock_class+0x14d0/0x14d0
[   29.562045][ T1046]  ? find_held_lock+0x39/0x1d0
[   29.562641][ T1046]  lock_acquire+0x164/0x3b0
[   29.563199][ T1046]  ? dev_mc_sync_multiple+0x95/0x120
[   29.563872][ T1046]  _raw_spin_lock+0x30/0x70
[   29.564464][ T1046]  ? dev_mc_sync_multiple+0x95/0x120
[   29.565146][ T1046]  dev_mc_sync_multiple+0x95/0x120
[   29.565793][ T1046]  bond_enslave+0x448d/0x47b0 [bonding]
[   29.566487][ T1046]  ? bond_update_slave_arr+0x940/0x940 [bonding]
[   29.567279][ T1046]  ? bstr_printf+0xc20/0xc20
[   29.567857][ T1046]  ? stack_trace_consume_entry+0x160/0x160
[   29.568614][ T1046]  ? deactivate_slab.isra.77+0x2c5/0x800
[   29.569320][ T1046]  ? check_chain_key+0x236/0x5d0
[   29.569939][ T1046]  ? sscanf+0x93/0xc0
[   29.570442][ T1046]  ? vsscanf+0x1e20/0x1e20
[   29.571003][ T1046]  bond_option_slaves_set+0x1a3/0x370 [bonding]
[ ... ]

Fixes: ab92d68fc22f ("net: core: add generic lockdep keys")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoNFC: pn544: Fix a typo in a debug message
Christophe JAILLET [Sat, 15 Feb 2020 07:17:28 +0000 (08:17 +0100)]
NFC: pn544: Fix a typo in a debug message

The ending character of the string shoulb be \n, not \b.

Fixes: 17936b43f0fd ("NFC: Standardize logging style")
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoethtool: fix application of verbose no_mask bitset
Michal Kubecek [Sat, 15 Feb 2020 00:55:53 +0000 (01:55 +0100)]
ethtool: fix application of verbose no_mask bitset

A bitset without mask in a _SET request means we want exactly the bits in
the bitset to be set. This works correctly for compact format but when
verbose format is parsed, ethnl_update_bitset32_verbose() only sets the
bits present in the request bitset but does not clear the rest. This can
cause incorrect results like

  lion:~ # ethtool eth0 | grep Wake
          Supports Wake-on: pumbg
          Wake-on: g
  lion:~ # ethtool -s eth0 wol u
  lion:~ # ethtool eth0 | grep Wake
          Supports Wake-on: pumbg
          Wake-on: ug

when the second ethtool command issues request

ETHTOOL_MSG_WOL_SET
    ETHTOOL_A_WOL_HEADER
        ETHTOOL_A_HEADER_DEV_NAME = "eth0"
    ETHTOOL_A_WOL_MODES
        ETHTOOL_A_BITSET_NOMASK
        ETHTOOL_A_BITSET_BITS
            ETHTOOL_A_BITSET_BITS_BIT
                ETHTOOL_BITSET_BIT_INDEX = 1

Fix the logic by clearing the whole target bitmap before we start iterating
through the request bits.

Fixes: 10b518d4e6dd ("ethtool: netlink bitset handling")
Signed-off-by: Michal Kubecek <mkubecek@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: b53: Ensure the default VID is untagged
Florian Fainelli [Fri, 14 Feb 2020 23:26:19 +0000 (15:26 -0800)]
net: dsa: b53: Ensure the default VID is untagged

We need to ensure that the default VID is untagged otherwise the switch
will be sending tagged frames and the results can be problematic. This
is especially true with b53 switches that use VID 0 as their default
VLAN since VID 0 has a special meaning.

Fixes: fea83353177a ("net: dsa: b53: Fix default VLAN ID")
Fixes: 061f6a505ac3 ("net: dsa: Add ndo_vlan_rx_{add, kill}_vid implementation")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'wireguard-fixes'
David S. Miller [Mon, 17 Feb 2020 03:21:56 +0000 (19:21 -0800)]
Merge branch 'wireguard-fixes'

Jason A. Donenfeld says:

====================
wireguard fixes for 5.6-rc2

Here are four fixes for wireguard collected since rc1:

1) Some small cleanups to the test suite to help massively parallel
   builds.

2) A change in how we reset our load calculation to avoid a more
   expensive comparison, suggested by Matt Dunwoodie.

3) I've been loading more and more of wireguard's surface into
   syzkaller, trying to get our coverage as complete as possible,
   leading in this case to a fix for mtu=0 devices.

4) A removal of superfluous code, pointed out by Eric Dumazet.

v2 fixes a logical problem in the patch for (3) pointed out by Eric Dumazet. v3
replaces some non-obvious bitmath in (3) with a more obvious expression, and
adds patch (4).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agowireguard: socket: remove extra call to synchronize_net
Jason A. Donenfeld [Fri, 14 Feb 2020 22:57:23 +0000 (23:57 +0100)]
wireguard: socket: remove extra call to synchronize_net

synchronize_net() is a wrapper around synchronize_rcu(), so there's no
point in having synchronize_net and synchronize_rcu back to back,
despite the documentation comment suggesting maybe it's somewhat useful,
"Wait for packets currently being received to be done." This commit
removes the extra call.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agowireguard: send: account for mtu=0 devices
Jason A. Donenfeld [Fri, 14 Feb 2020 22:57:22 +0000 (23:57 +0100)]
wireguard: send: account for mtu=0 devices

It turns out there's an easy way to get packets queued up while still
having an MTU of zero, and that's via persistent keep alive. This commit
makes sure that in whatever condition, we don't wind up dividing by
zero. Note that an MTU of zero for a wireguard interface is something
quasi-valid, so I don't think the correct fix is to limit it via
min_mtu. This can be reproduced easily with:

ip link add wg0 type wireguard
ip link add wg1 type wireguard
ip link set wg0 up mtu 0
ip link set wg1 up
wg set wg0 private-key <(wg genkey)
wg set wg1 listen-port 1 private-key <(wg genkey) peer $(wg show wg0 public-key)
wg set wg0 peer $(wg show wg1 public-key) persistent-keepalive 1 endpoint 127.0.0.1:1

However, while min_mtu=0 seems fine, it makes sense to restrict the
max_mtu. This commit also restricts the maximum MTU to the greatest
number for which rounding up to the padding multiple won't overflow a
signed integer. Packets this large were always rejected anyway
eventually, due to checks deeper in, but it seems more sound not to even
let the administrator configure something that won't work anyway.

We use this opportunity to clean up this function a bit so that it's
clear which paths we're expecting.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agowireguard: receive: reset last_under_load to zero
Jason A. Donenfeld [Fri, 14 Feb 2020 22:57:21 +0000 (23:57 +0100)]
wireguard: receive: reset last_under_load to zero

This is a small optimization that prevents more expensive comparisons
from happening when they are no longer necessary, by clearing the
last_under_load variable whenever we wind up in a state where we were
under load but we no longer are.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Suggested-by: Matt Dunwoodie <ncon@noconroy.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agowireguard: selftests: reduce complexity and fix make races
Jason A. Donenfeld [Fri, 14 Feb 2020 22:57:20 +0000 (23:57 +0100)]
wireguard: selftests: reduce complexity and fix make races

This gives us fewer dependencies and shortens build time, fixes up some
hash checking race conditions, and also fixes missing directory creation
that caused issues on massively parallel builds.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: Protect subflow socket options before connection completes
Mat Martineau [Fri, 14 Feb 2020 22:14:29 +0000 (14:14 -0800)]
mptcp: Protect subflow socket options before connection completes

Userspace should not be able to directly manipulate subflow socket
options before a connection is established since it is not yet known if
it will be an MPTCP subflow or a TCP fallback subflow. TCP fallback
subflows can be more directly controlled by userspace because they are
regular TCP connections, while MPTCP subflow sockets need to be
configured for the specific needs of MPTCP. Use the same logic as
sendmsg/recvmsg to ensure that socket option calls are only passed
through to known TCP fallback subflows.

Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: restore mdio regs in the iproc mdio driver
Arun Parameswaran [Fri, 14 Feb 2020 21:47:46 +0000 (13:47 -0800)]
net: phy: restore mdio regs in the iproc mdio driver

The mii management register in iproc mdio block
does not have a retention register so it is lost on suspend.
Save and restore value of register while resuming from suspend.

Fixes: bb1a619735b4 ("net: phy: Initialize mdio clock at probe function")
Signed-off-by: Arun Parameswaran <arun.parameswaran@broadcom.com>
Signed-off-by: Scott Branden <scott.branden@broadcom.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: add strict checks in netdev_name_node_alt_destroy()
Eric Dumazet [Fri, 14 Feb 2020 15:53:53 +0000 (07:53 -0800)]
net: add strict checks in netdev_name_node_alt_destroy()

netdev_name_node_alt_destroy() does a lookup over all
device names of a namespace.

We need to make sure the name belongs to the device
of interest, and that we do not destroy its primary
name, since we rely on it being not deleted :
dev->name_node would indeed point to freed memory.

syzbot report was the following :

BUG: KASAN: use-after-free in dev_net include/linux/netdevice.h:2206 [inline]
BUG: KASAN: use-after-free in mld_force_mld_version net/ipv6/mcast.c:1172 [inline]
BUG: KASAN: use-after-free in mld_in_v2_mode_only net/ipv6/mcast.c:1180 [inline]
BUG: KASAN: use-after-free in mld_in_v1_mode+0x203/0x230 net/ipv6/mcast.c:1190
Read of size 8 at addr ffff88809886c588 by task swapper/1/0

CPU: 1 PID: 0 Comm: swapper/1 Not tainted 5.6.0-rc1-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 <IRQ>
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x197/0x210 lib/dump_stack.c:118
 print_address_description.constprop.0.cold+0xd4/0x30b mm/kasan/report.c:374
 __kasan_report.cold+0x1b/0x32 mm/kasan/report.c:506
 kasan_report+0x12/0x20 mm/kasan/common.c:641
 __asan_report_load8_noabort+0x14/0x20 mm/kasan/generic_report.c:135
 dev_net include/linux/netdevice.h:2206 [inline]
 mld_force_mld_version net/ipv6/mcast.c:1172 [inline]
 mld_in_v2_mode_only net/ipv6/mcast.c:1180 [inline]
 mld_in_v1_mode+0x203/0x230 net/ipv6/mcast.c:1190
 mld_send_initial_cr net/ipv6/mcast.c:2083 [inline]
 mld_dad_timer_expire+0x24/0x230 net/ipv6/mcast.c:2118
 call_timer_fn+0x1ac/0x780 kernel/time/timer.c:1404
 expire_timers kernel/time/timer.c:1449 [inline]
 __run_timers kernel/time/timer.c:1773 [inline]
 __run_timers kernel/time/timer.c:1740 [inline]
 run_timer_softirq+0x6c3/0x1790 kernel/time/timer.c:1786
 __do_softirq+0x262/0x98c kernel/softirq.c:292
 invoke_softirq kernel/softirq.c:373 [inline]
 irq_exit+0x19b/0x1e0 kernel/softirq.c:413
 exiting_irq arch/x86/include/asm/apic.h:546 [inline]
 smp_apic_timer_interrupt+0x1a3/0x610 arch/x86/kernel/apic/apic.c:1146
 apic_timer_interrupt+0xf/0x20 arch/x86/entry/entry_64.S:829
 </IRQ>
RIP: 0010:native_safe_halt+0xe/0x10 arch/x86/include/asm/irqflags.h:61
Code: 68 73 c5 f9 eb 8a cc cc cc cc cc cc e9 07 00 00 00 0f 00 2d 94 be 59 00 f4 c3 66 90 e9 07 00 00 00 0f 00 2d 84 be 59 00 fb f4 <c3> cc 55 48 89 e5 41 57 41 56 41 55 41 54 53 e8 de 2a 74 f9 e8 09
RSP: 0018:ffffc90000d3fd68 EFLAGS: 00000282 ORIG_RAX: ffffffffffffff13
RAX: 1ffffffff136761a RBX: ffff8880a99fc340 RCX: 0000000000000000
RDX: dffffc0000000000 RSI: 0000000000000006 RDI: ffff8880a99fcbd4
RBP: ffffc90000d3fd98 R08: ffff8880a99fc340 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000000 R12: dffffc0000000000
R13: ffffffff8aa5a1c0 R14: 0000000000000000 R15: 0000000000000001
 arch_cpu_idle+0xa/0x10 arch/x86/kernel/process.c:686
 default_idle_call+0x84/0xb0 kernel/sched/idle.c:94
 cpuidle_idle_call kernel/sched/idle.c:154 [inline]
 do_idle+0x3c8/0x6e0 kernel/sched/idle.c:269
 cpu_startup_entry+0x1b/0x20 kernel/sched/idle.c:361
 start_secondary+0x2f4/0x410 arch/x86/kernel/smpboot.c:264
 secondary_startup_64+0xa4/0xb0 arch/x86/kernel/head_64.S:242

Allocated by task 10229:
 save_stack+0x23/0x90 mm/kasan/common.c:72
 set_track mm/kasan/common.c:80 [inline]
 __kasan_kmalloc mm/kasan/common.c:515 [inline]
 __kasan_kmalloc.constprop.0+0xcf/0xe0 mm/kasan/common.c:488
 kasan_kmalloc+0x9/0x10 mm/kasan/common.c:529
 __do_kmalloc_node mm/slab.c:3616 [inline]
 __kmalloc_node+0x4e/0x70 mm/slab.c:3623
 kmalloc_node include/linux/slab.h:578 [inline]
 kvmalloc_node+0x68/0x100 mm/util.c:574
 kvmalloc include/linux/mm.h:645 [inline]
 kvzalloc include/linux/mm.h:653 [inline]
 alloc_netdev_mqs+0x98/0xe40 net/core/dev.c:9797
 rtnl_create_link+0x22d/0xaf0 net/core/rtnetlink.c:3047
 __rtnl_newlink+0xf9f/0x1790 net/core/rtnetlink.c:3309
 rtnl_newlink+0x69/0xa0 net/core/rtnetlink.c:3377
 rtnetlink_rcv_msg+0x45e/0xaf0 net/core/rtnetlink.c:5438
 netlink_rcv_skb+0x177/0x450 net/netlink/af_netlink.c:2477
 rtnetlink_rcv+0x1d/0x30 net/core/rtnetlink.c:5456
 netlink_unicast_kernel net/netlink/af_netlink.c:1302 [inline]
 netlink_unicast+0x59e/0x7e0 net/netlink/af_netlink.c:1328
 netlink_sendmsg+0x91c/0xea0 net/netlink/af_netlink.c:1917
 sock_sendmsg_nosec net/socket.c:652 [inline]
 sock_sendmsg+0xd7/0x130 net/socket.c:672
 __sys_sendto+0x262/0x380 net/socket.c:1998
 __do_compat_sys_socketcall net/compat.c:771 [inline]
 __se_compat_sys_socketcall net/compat.c:719 [inline]
 __ia32_compat_sys_socketcall+0x530/0x710 net/compat.c:719
 do_syscall_32_irqs_on arch/x86/entry/common.c:337 [inline]
 do_fast_syscall_32+0x27b/0xe16 arch/x86/entry/common.c:408
 entry_SYSENTER_compat+0x70/0x7f arch/x86/entry/entry_64_compat.S:139

Freed by task 10229:
 save_stack+0x23/0x90 mm/kasan/common.c:72
 set_track mm/kasan/common.c:80 [inline]
 kasan_set_free_info mm/kasan/common.c:337 [inline]
 __kasan_slab_free+0x102/0x150 mm/kasan/common.c:476
 kasan_slab_free+0xe/0x10 mm/kasan/common.c:485
 __cache_free mm/slab.c:3426 [inline]
 kfree+0x10a/0x2c0 mm/slab.c:3757
 __netdev_name_node_alt_destroy+0x1ff/0x2a0 net/core/dev.c:322
 netdev_name_node_alt_destroy+0x57/0x80 net/core/dev.c:334
 rtnl_alt_ifname net/core/rtnetlink.c:3518 [inline]
 rtnl_linkprop.isra.0+0x575/0x6f0 net/core/rtnetlink.c:3567
 rtnl_dellinkprop+0x46/0x60 net/core/rtnetlink.c:3588
 rtnetlink_rcv_msg+0x45e/0xaf0 net/core/rtnetlink.c:5438
 netlink_rcv_skb+0x177/0x450 net/netlink/af_netlink.c:2477
 rtnetlink_rcv+0x1d/0x30 net/core/rtnetlink.c:5456
 netlink_unicast_kernel net/netlink/af_netlink.c:1302 [inline]
 netlink_unicast+0x59e/0x7e0 net/netlink/af_netlink.c:1328
 netlink_sendmsg+0x91c/0xea0 net/netlink/af_netlink.c:1917
 sock_sendmsg_nosec net/socket.c:652 [inline]
 sock_sendmsg+0xd7/0x130 net/socket.c:672
 ____sys_sendmsg+0x753/0x880 net/socket.c:2343
 ___sys_sendmsg+0x100/0x170 net/socket.c:2397
 __sys_sendmsg+0x105/0x1d0 net/socket.c:2430
 __compat_sys_sendmsg net/compat.c:642 [inline]
 __do_compat_sys_sendmsg net/compat.c:649 [inline]
 __se_compat_sys_sendmsg net/compat.c:646 [inline]
 __ia32_compat_sys_sendmsg+0x7a/0xb0 net/compat.c:646
 do_syscall_32_irqs_on arch/x86/entry/common.c:337 [inline]
 do_fast_syscall_32+0x27b/0xe16 arch/x86/entry/common.c:408
 entry_SYSENTER_compat+0x70/0x7f arch/x86/entry/entry_64_compat.S:139

The buggy address belongs to the object at ffff88809886c000
 which belongs to the cache kmalloc-4k of size 4096
The buggy address is located 1416 bytes inside of
 4096-byte region [ffff88809886c000ffff88809886d000)
The buggy address belongs to the page:
page:ffffea0002621b00 refcount:1 mapcount:0 mapping:ffff8880aa402000 index:0x0 compound_mapcount: 0
flags: 0xfffe0000010200(slab|head)
raw: 00fffe0000010200 ffffea0002610d08 ffffea0002607608 ffff8880aa402000
raw: 0000000000000000 ffff88809886c000 0000000100000001 0000000000000000
page dumped because: kasan: bad access detected

Memory state around the buggy address:
 ffff88809886c480: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
 ffff88809886c500: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
>ffff88809886c580: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
                      ^
 ffff88809886c600: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
 ffff88809886c680: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb

Fixes: 36fbf1e52bd3 ("net: rtnetlink: add linkprop commands to add and delete alternative ifnames")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'atlantic-fixes'
David S. Miller [Mon, 17 Feb 2020 03:03:40 +0000 (19:03 -0800)]
Merge branch 'atlantic-fixes'

Igor Russkikh says:

====================
Marvell atlantic 2020/02 updates

Hi David, here is another set of bugfixes on AQC family found on
last integration phase.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: atlantic: fix out of range usage of active_vlans array
Dmitry Bogdanov [Fri, 14 Feb 2020 15:44:58 +0000 (18:44 +0300)]
net: atlantic: fix out of range usage of active_vlans array

fix static checker warning:
 drivers/net/ethernet/aquantia/atlantic/aq_filters.c:166 aq_check_approve_fvlan()
 error: passing untrusted data to 'test_bit()'

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Fixes: 7975d2aff5af: ("net: aquantia: add support of rx-vlan-filter offload")
Signed-off-by: Dmitry Bogdanov <dbogdanov@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: atlantic: possible fault in transition to hibernation
Pavel Belous [Fri, 14 Feb 2020 15:44:57 +0000 (18:44 +0300)]
net: atlantic: possible fault in transition to hibernation

during hibernation freeze, aq_nic_stop could be invoked
on a stopped device. That may cause panic on access to
not yet allocated vector/ring structures.

Add a check to stop device if it is not yet stopped.

Similiarly after freeze in hibernation thaw, aq_nic_start
could be invoked on a not initialized net device.
Result will be the same.

Add a check to start device if it is initialized.
In our case, this is the same as started.

Fixes: 8aaa112a57c1 ("net: atlantic: refactoring pm logic")
Signed-off-by: Pavel Belous <pbelous@marvell.com>
Signed-off-by: Nikita Danilov <ndanilov@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: Dmitry Bogdanov <dbogdanov@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: atlantic: fix potential error handling
Pavel Belous [Fri, 14 Feb 2020 15:44:56 +0000 (18:44 +0300)]
net: atlantic: fix potential error handling

Code inspection found that in case of mapping error we do return current
'ret' value. But beside error, it is used to count number of descriptors
allocated for the packet. In that case map_skb function could return '1'.

Changing it to return zero (number of mapped descriptors for skb)

Fixes: 018423e90bee ("net: ethernet: aquantia: Add ring support code")
Signed-off-by: Pavel Belous <pbelous@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: Dmitry Bogdanov <dbogdanov@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: atlantic: fix use after free kasan warn
Pavel Belous [Fri, 14 Feb 2020 15:44:55 +0000 (18:44 +0300)]
net: atlantic: fix use after free kasan warn

skb->len is used to calculate statistics after xmit invocation.

Under a stress load it may happen that skb will be xmited,
rx interrupt will come and skb will be freed, all before xmit function
is even returned.

Eventually, skb->len will access unallocated area.

Moving stats calculation into tx_clean routine.

Fixes: 018423e90bee ("net: ethernet: aquantia: Add ring support code")
Reported-by: Christophe Vu-Brugier <cvubrugier@fastmail.fm>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: Pavel Belous <pbelous@marvell.com>
Signed-off-by: Dmitry Bogdanov <dbogdanov@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: atlantic: better loopback mode handling
Nikita Danilov [Fri, 14 Feb 2020 15:44:54 +0000 (18:44 +0300)]
net: atlantic: better loopback mode handling

Add checks to not enable multiple loopback modes simultaneously,
It was also discovered that for dma loopback to function correctly
promisc mode should be enabled on device.

Fixes: ea4b4d7fc106 ("net: atlantic: loopback tests via private flags")
Signed-off-by: Nikita Danilov <ndanilov@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: Dmitry Bogdanov <dbogdanov@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: atlantic: ptp gpio adjustments
Egor Pomozov [Fri, 14 Feb 2020 15:44:53 +0000 (18:44 +0300)]
net: atlantic: ptp gpio adjustments

Clock adjustment data should be passed to FW as well, otherwise in some
cases a drift was observed when using GPIO features.

Signed-off-by: Egor Pomozov <epomozov@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: Dmitry Bogdanov <dbogdanov@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: atlantic: check rpc result and wait for rpc address
Igor Russkikh [Fri, 14 Feb 2020 15:44:52 +0000 (18:44 +0300)]
net: atlantic: check rpc result and wait for rpc address

Artificial HW reliability tests revealed a possible hangup in
the driver. Normally, when device disappears from bus, all
register reads returns 0xFFFFFFFF.

At remote procedure invocation towards FW there is a logic
where result is compared with -1 in a loop.
That caused an infinite loop if hardware due to some issues
disappears from bus.

Add extra result checks to prevent this.

Signed-off-by: Dmitry Bogdanov <dbogdanov@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: atlantic: checksum compat issue
Dmitry Bezrukov [Fri, 14 Feb 2020 15:44:51 +0000 (18:44 +0300)]
net: atlantic: checksum compat issue

Yet another checksum offload compatibility issue was found.

The known issue is that AQC HW marks tcp packets with 0xFFFF checksum
as invalid (1). This is workarounded in driver, passing all the suspicious
packets up to the stack for further csum validation.

Another HW problem (2) is that it hides invalid csum of LRO aggregated
packets inside of the individual descriptors. That was workarounded
by forced scan of all LRO descriptors for checksum errors.

However the scan logic was joint for both LRO and multi-descriptor
packets (jumbos). And this causes the issue.

We have to drop LRO packets with the detected bad checksum
because of (2), but we have to pass jumbo packets to stack because of (1).

When using windows tcp partner with jumbo frames but with LSO disabled
driver discards such frames as bad checksummed. But only LRO frames
should be dropped, not jumbos.

On such a configurations tcp stream have a chance of drops and stucks.

(1) 76f254d4afe2 ("net: aquantia: tcp checksum 0xffff being handled incorrectly")
(2) d08b9a0a3ebd ("net: aquantia: do not pass lro session with invalid tcp checksum")

Fixes: d08b9a0a3ebd ("net: aquantia: do not pass lro session with invalid tcp checksum")
Signed-off-by: Dmitry Bezrukov <dbezrukov@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: Dmitry Bogdanov <dbogdanov@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: forwarding: vxlan_bridge_1d: fix tos value
Hangbin Liu [Thu, 13 Feb 2020 09:40:54 +0000 (17:40 +0800)]
selftests: forwarding: vxlan_bridge_1d: fix tos value

After commit 71130f29979c ("vxlan: fix tos value before xmit") we start
strict vxlan xmit tos value by RT_TOS(), which limits the tos value less
than 0x1E. With current value 0x40 the test will failed with "v1: Expected
to capture 10 packets, got 0". So let's choose a smaller tos value for
testing.

Fixes: d417ecf533fe ("selftests: forwarding: vxlan_bridge_1d: Add a TOS test")
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: rtnetlink: fix bugs in rtnl_alt_ifname()
Eric Dumazet [Thu, 13 Feb 2020 04:58:26 +0000 (20:58 -0800)]
net: rtnetlink: fix bugs in rtnl_alt_ifname()

Since IFLA_ALT_IFNAME is an NLA_STRING, we have no
guarantee it is nul terminated.

We should use nla_strdup() instead of kstrdup(), since this
helper will make sure not accessing out-of-bounds data.

BUG: KMSAN: uninit-value in strlen+0x5e/0xa0 lib/string.c:535
CPU: 1 PID: 19157 Comm: syz-executor.5 Not tainted 5.5.0-rc5-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x1c9/0x220 lib/dump_stack.c:118
 kmsan_report+0xf7/0x1e0 mm/kmsan/kmsan_report.c:118
 __msan_warning+0x58/0xa0 mm/kmsan/kmsan_instr.c:215
 strlen+0x5e/0xa0 lib/string.c:535
 kstrdup+0x7f/0x1a0 mm/util.c:59
 rtnl_alt_ifname net/core/rtnetlink.c:3495 [inline]
 rtnl_linkprop+0x85d/0xc00 net/core/rtnetlink.c:3553
 rtnl_newlinkprop+0x9d/0xb0 net/core/rtnetlink.c:3568
 rtnetlink_rcv_msg+0x1153/0x1570 net/core/rtnetlink.c:5424
 netlink_rcv_skb+0x451/0x650 net/netlink/af_netlink.c:2477
 rtnetlink_rcv+0x50/0x60 net/core/rtnetlink.c:5442
 netlink_unicast_kernel net/netlink/af_netlink.c:1302 [inline]
 netlink_unicast+0xf9e/0x1100 net/netlink/af_netlink.c:1328
 netlink_sendmsg+0x1248/0x14d0 net/netlink/af_netlink.c:1917
 sock_sendmsg_nosec net/socket.c:639 [inline]
 sock_sendmsg net/socket.c:659 [inline]
 ____sys_sendmsg+0x12b6/0x1350 net/socket.c:2330
 ___sys_sendmsg net/socket.c:2384 [inline]
 __sys_sendmsg+0x451/0x5f0 net/socket.c:2417
 __do_sys_sendmsg net/socket.c:2426 [inline]
 __se_sys_sendmsg+0x97/0xb0 net/socket.c:2424
 __x64_sys_sendmsg+0x4a/0x70 net/socket.c:2424
 do_syscall_64+0xb8/0x160 arch/x86/entry/common.c:296
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x45b3b9
Code: ad b6 fb ff c3 66 2e 0f 1f 84 00 00 00 00 00 66 90 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 7b b6 fb ff c3 66 2e 0f 1f 84 00 00 00 00
RSP: 002b:00007ff1c7b1ac78 EFLAGS: 00000246 ORIG_RAX: 000000000000002e
RAX: ffffffffffffffda RBX: 00007ff1c7b1b6d4 RCX: 000000000045b3b9
RDX: 0000000000000000 RSI: 0000000020000040 RDI: 0000000000000003
RBP: 000000000075bf20 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000246 R12: 00000000ffffffff
R13: 00000000000009cb R14: 00000000004cb3dd R15: 000000000075bf2c

Uninit was created at:
 kmsan_save_stack_with_flags mm/kmsan/kmsan.c:144 [inline]
 kmsan_internal_poison_shadow+0x66/0xd0 mm/kmsan/kmsan.c:127
 kmsan_slab_alloc+0x8a/0xe0 mm/kmsan/kmsan_hooks.c:82
 slab_alloc_node mm/slub.c:2774 [inline]
 __kmalloc_node_track_caller+0xb40/0x1200 mm/slub.c:4382
 __kmalloc_reserve net/core/skbuff.c:141 [inline]
 __alloc_skb+0x2fd/0xac0 net/core/skbuff.c:209
 alloc_skb include/linux/skbuff.h:1049 [inline]
 netlink_alloc_large_skb net/netlink/af_netlink.c:1174 [inline]
 netlink_sendmsg+0x7d3/0x14d0 net/netlink/af_netlink.c:1892
 sock_sendmsg_nosec net/socket.c:639 [inline]
 sock_sendmsg net/socket.c:659 [inline]
 ____sys_sendmsg+0x12b6/0x1350 net/socket.c:2330
 ___sys_sendmsg net/socket.c:2384 [inline]
 __sys_sendmsg+0x451/0x5f0 net/socket.c:2417
 __do_sys_sendmsg net/socket.c:2426 [inline]
 __se_sys_sendmsg+0x97/0xb0 net/socket.c:2424
 __x64_sys_sendmsg+0x4a/0x70 net/socket.c:2424
 do_syscall_64+0xb8/0x160 arch/x86/entry/common.c:296
 entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fixes: 36fbf1e52bd3 ("net: rtnetlink: add linkprop commands to add and delete alternative ifnames")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jiri Pirko <jiri@mellanox.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: macb: ensure interface is not suspended on at91rm9200
Alexandre Belloni [Wed, 12 Feb 2020 16:45:38 +0000 (17:45 +0100)]
net: macb: ensure interface is not suspended on at91rm9200

Because of autosuspend, at91ether_start is called with clocks disabled.
Ensure that pm_runtime doesn't suspend the interface as soon as it is
opened as there is no pm_runtime support is the other relevant parts of the
platform support for at91rm9200.

Fixes: d54f89af6cc4 ("net: macb: Add pm runtime support")
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Reviewed-by: Claudiu Beznea <claudiu.beznea@microchip.com>
Acked-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: fib_rules: Correctly set table field when table number exceeds 8 bits
Jethro Beekman [Wed, 12 Feb 2020 15:43:41 +0000 (16:43 +0100)]
net: fib_rules: Correctly set table field when table number exceeds 8 bits

In 709772e6e06564ed94ba740de70185ac3d792773, RT_TABLE_COMPAT was added to
allow legacy software to deal with routing table numbers >= 256, but the
same change to FIB rule queries was overlooked.

Signed-off-by: Jethro Beekman <jethro@fortanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/rds: Track user mapped pages through special API
Leon Romanovsky [Wed, 12 Feb 2020 03:03:55 +0000 (19:03 -0800)]
net/rds: Track user mapped pages through special API

Convert net/rds to use the newly introduces pin_user_pages() API,
which properly sets FOLL_PIN. Setting FOLL_PIN is now required for
code that requires tracking of pinned pages.

Note that this effectively changes the code's behavior: it now
ultimately calls set_page_dirty_lock(), instead of set_page_dirty().
This is probably more accurate.

As Christoph Hellwig put it, "set_page_dirty() is only safe if we are
dealing with a file backed page where we have reference on the inode it
hangs off." [1]

[1] https://lore.kernel.org/r/20190723153640.GB720@lst.de

Cc: Hans Westgaard Ry <hans.westgaard.ry@oracle.com>
Cc: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoipv6: Fix nlmsg_flags when splitting a multipath route
Benjamin Poirier [Wed, 12 Feb 2020 01:41:07 +0000 (10:41 +0900)]
ipv6: Fix nlmsg_flags when splitting a multipath route

When splitting an RTA_MULTIPATH request into multiple routes and adding the
second and later components, we must not simply remove NLM_F_REPLACE but
instead replace it by NLM_F_CREATE. Otherwise, it may look like the netlink
message was malformed.

For example,
ip route add 2001:db8::1/128 dev dummy0
ip route change 2001:db8::1/128 nexthop via fe80::30:1 dev dummy0 \
nexthop via fe80::30:2 dev dummy0
results in the following warnings:
[ 1035.057019] IPv6: RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE
[ 1035.057517] IPv6: NLM_F_CREATE should be set when creating new route

This patch makes the nlmsg sequence look equivalent for __ip6_ins_rt() to
what it would get if the multipath route had been added in multiple netlink
operations:
ip route add 2001:db8::1/128 dev dummy0
ip route change 2001:db8::1/128 nexthop via fe80::30:1 dev dummy0
ip route append 2001:db8::1/128 nexthop via fe80::30:2 dev dummy0

Fixes: 27596472473a ("ipv6: fix ECMP route replacement")
Signed-off-by: Benjamin Poirier <bpoirier@cumulusnetworks.com>
Reviewed-by: Michal Kubecek <mkubecek@suse.cz>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoipv6: Fix route replacement with dev-only route
Benjamin Poirier [Wed, 12 Feb 2020 01:41:06 +0000 (10:41 +0900)]
ipv6: Fix route replacement with dev-only route

After commit 27596472473a ("ipv6: fix ECMP route replacement") it is no
longer possible to replace an ECMP-able route by a non ECMP-able route.
For example,
ip route add 2001:db8::1/128 via fe80::1 dev dummy0
ip route replace 2001:db8::1/128 dev dummy0
does not work as expected.

Tweak the replacement logic so that point 3 in the log of the above commit
becomes:
3. If the new route is not ECMP-able, and no matching non-ECMP-able route
exists, replace matching ECMP-able route (if any) or add the new route.

We can now summarize the entire replace semantics to:
When doing a replace, prefer replacing a matching route of the same
"ECMP-able-ness" as the replace argument. If there is no such candidate,
fallback to the first route found.

Fixes: 27596472473a ("ipv6: fix ECMP route replacement")
Signed-off-by: Benjamin Poirier <bpoirier@cumulusnetworks.com>
Reviewed-by: Michal Kubecek <mkubecek@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: forwarding: use proto icmp for {gretap, ip6gretap}_mac testing
Hangbin Liu [Tue, 11 Feb 2020 07:32:56 +0000 (15:32 +0800)]
selftests: forwarding: use proto icmp for {gretap, ip6gretap}_mac testing

For tc ip_proto filter, when we extract the flow via __skb_flow_dissect()
without flag FLOW_DISSECTOR_F_STOP_AT_ENCAP, we will continue extract to
the inner proto.

So for GRE + ICMP messages, we should not track GRE proto, but inner ICMP
proto.

For test mirror_gre.sh, it may make user confused if we capture ICMP
message on $h3(since the flow is GRE message). So I move the capture
dev to h3-gt{4,6}, and only capture ICMP message.

Before the fix:
]# ./mirror_gre.sh
TEST: ingress mirror to gretap (skip_hw)                            [ OK ]
TEST: egress mirror to gretap (skip_hw)                             [ OK ]
TEST: ingress mirror to ip6gretap (skip_hw)                         [ OK ]
TEST: egress mirror to ip6gretap (skip_hw)                          [ OK ]
TEST: ingress mirror to gretap: envelope MAC (skip_hw)              [FAIL]
 Expected to capture 10 packets, got 0.
TEST: egress mirror to gretap: envelope MAC (skip_hw)               [FAIL]
 Expected to capture 10 packets, got 0.
TEST: ingress mirror to ip6gretap: envelope MAC (skip_hw)           [FAIL]
 Expected to capture 10 packets, got 0.
TEST: egress mirror to ip6gretap: envelope MAC (skip_hw)            [FAIL]
 Expected to capture 10 packets, got 0.
TEST: two simultaneously configured mirrors (skip_hw)               [ OK ]
WARN: Could not test offloaded functionality

After fix:
]# ./mirror_gre.sh
TEST: ingress mirror to gretap (skip_hw)                            [ OK ]
TEST: egress mirror to gretap (skip_hw)                             [ OK ]
TEST: ingress mirror to ip6gretap (skip_hw)                         [ OK ]
TEST: egress mirror to ip6gretap (skip_hw)                          [ OK ]
TEST: ingress mirror to gretap: envelope MAC (skip_hw)              [ OK ]
TEST: egress mirror to gretap: envelope MAC (skip_hw)               [ OK ]
TEST: ingress mirror to ip6gretap: envelope MAC (skip_hw)           [ OK ]
TEST: egress mirror to ip6gretap: envelope MAC (skip_hw)            [ OK ]
TEST: two simultaneously configured mirrors (skip_hw)               [ OK ]
WARN: Could not test offloaded functionality

Fixes: ba8d39871a10 ("selftests: forwarding: Add test for mirror to gretap")
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Petr Machata <pmachata@gmail.com>
Tested-by: Petr Machata <pmachata@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Fri, 14 Feb 2020 20:40:38 +0000 (12:40 -0800)]
Merge git://git./linux/kernel/git/netdev/net

Pull networking fixes from David Miller:

 1) Fix interrupt name truncation in mv88e6xxx dsa driver, from Andrew
    Lunn.

 2) Process generic XDP even if SKB is cloned, from Toke Høiland-Jørgensen.

 3) Fix leak of kernel memory to userspace in smc, from Eric Dumazet.

 4) Add some missing netlink attribute validation to matchall and
    flower, from Davide Caratti.

 5) Send icmp responses properly when NAT has been applied to the frame
    before we get to the tunnel emitting the icmp, from Jason Donenfeld.

 6) Make sure there is enough SKB headroom when adding dsa tags for qca
    and ar9331. From Per Forlin.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (62 commits)
  netdevice.h: fix all kernel-doc and Sphinx warnings
  net: dsa: tag_ar9331: Make sure there is headroom for tag
  net: dsa: tag_qca: Make sure there is headroom for tag
  net, ip6_tunnel: enhance tunnel locate with link check
  net/smc: no peer ID in CLC decline for SMCD
  net/smc: transfer fasync_list in case of fallback
  net: hns3: fix a copying IPv6 address error in hclge_fd_get_flow_tuples()
  net: hns3: fix VF bandwidth does not take effect in some case
  net: hns3: add management table after IMP reset
  mac80211: fix wrong 160/80+80 MHz setting
  cfg80211: add missing policy for NL80211_ATTR_STATUS_CODE
  xfrm: interface: use icmp_ndo_send helper
  wireguard: device: use icmp_ndo_send helper
  sunvnet: use icmp_ndo_send helper
  gtp: use icmp_ndo_send helper
  icmp: introduce helper for nat'd source address in network device context
  net/sched: flower: add missing validation of TCA_FLOWER_FLAGS
  net/sched: matchall: add missing validation of TCA_MATCHALL_FLAGS
  net/flow_dissector: remove unexist field description
  page_pool: refill page when alloc.count of pool is zero
  ...

4 years agoMerge tag 'pm-5.6-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Linus Torvalds [Fri, 14 Feb 2020 20:34:30 +0000 (12:34 -0800)]
Merge tag 'pm-5.6-rc2' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management fixes from Rafael Wysocki:
 "Fix three issues related to the handling of wakeup events signaled
  through the ACPI SCI while suspended to idle (Rafael Wysocki) and
  unexport an internal cpufreq variable (Yangtao Li)"

* tag 'pm-5.6-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPI: PM: s2idle: Prevent spurious SCIs from waking up the system
  ACPICA: Introduce acpi_any_gpe_status_set()
  ACPI: PM: s2idle: Avoid possible race related to the EC GPE
  ACPI: EC: Fix flushing of pending work
  cpufreq: Make cpufreq_global_kobject static

4 years agoMerge tag 'sound-5.6-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
Linus Torvalds [Fri, 14 Feb 2020 20:27:54 +0000 (12:27 -0800)]
Merge tag 'sound-5.6-rc2' of git://git./linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "The only common change is the regression fix of the previous PCM fix
  patch for managed buffers while the rest are usual suspects, USB-audio
  and HD-audio device-specific quirks.

  The change for UAC2 clock validation workaround became a bit big, but
  the changes are fairly straightforward"

* tag 'sound-5.6-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ALSA: pcm: Fix double hw_free calls
  ALSA: usb-audio: Add clock validity quirk for Denon MC7000/MCX8000
  ALSA: hda/realtek - Fix silent output on MSI-GL73
  ALSA: hda/realtek - Add more codec supported Headset Button
  ALSA: usb-audio: Apply sample rate quirk for Audioengine D1
  ALSA: usb-audio: Fix UAC2/3 effect unit parsing
  ALSA: usb-audio: Apply 48kHz fixed rate playback for Jabra Evolve 65 headset

4 years agoMerge tag 'drm-fixes-2020-02-14' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Fri, 14 Feb 2020 20:23:16 +0000 (12:23 -0800)]
Merge tag 'drm-fixes-2020-02-14' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "The core has a build fix for edid code on certain compilers/arches/,
  one MST fix and one vgem fix. Regular amdgpu fixes, and a couple of
  small driver fixes.

  The i915 fixes are bit larger than normal for this stage, but they
  were having CI issues last week, and they hadn't sent any fixes last
  week due to this.

  core:
   - edid build fix

  mst:
   - fix NULL ptr deref

  vgem:
   - fix close after free

  msm:
   - better dma-api usage

  sun4i:
   - disable allow_fb_modifiers

  amdgpu:
   - Additional OD fixes for navi
   - Misc display fixes
   - VCN 2.5 DPG fix
   - Prevent build errors on PowerPC on some configs
   - GDS EDC fix

  i915:
   - dsi/acpi fixes
   - gvt locking and allocation fixes
   - gem/gt fixes
   - bios timing parameters fix"

* tag 'drm-fixes-2020-02-14' of git://anongit.freedesktop.org/drm/drm: (50 commits)
  drm/i915: Mark the removal of the i915_request from the sched.link
  drm/i915/execlists: Reclaim the hanging virtual request
  drm/i915/execlists: Take a reference while capturing the guilty request
  drm/i915/execlists: Offline error capture
  drm/i915/gt: Allow temporary suspension of inflight requests
  drm/i915: Keep track of request among the scheduling lists
  drm/i915/gem: Tighten checks and acquiring the mmap object
  drm/i915: Fix preallocated barrier list append
  drm/i915/gt: Acquire ce->active before ce->pin_count/ce->pin_mutex
  drm/i915: Tighten atomicity of i915_active_acquire vs i915_active_release
  drm/i915: Stub out i915_gpu_coredump_put
  drm/amdgpu:/navi10: use the ODCAP enum to index the caps array
  drm/amdgpu: update smu_v11_0_pptable.h
  drm/amdgpu: correct comment to clear up the confusion
  drm/amd/display: DCN2.x Do not program DPPCLK if same value
  drm/amd/display: Don't map ATOM_ENABLE to ATOM_INIT
  drm/amdgpu/vcn2.5: fix warning
  drm/amdgpu: limit GDS clearing workaround in cold boot sequence
  drm/amdgpu: fix amdgpu pmu to use hwc->config instead of hwc->conf
  amdgpu: Prevent build errors regarding soft/hard-float FP ABI tags
  ...

4 years agonetdevice.h: fix all kernel-doc and Sphinx warnings
Randy Dunlap [Thu, 13 Feb 2020 06:28:20 +0000 (22:28 -0800)]
netdevice.h: fix all kernel-doc and Sphinx warnings

Eliminate all kernel-doc and Sphinx warnings in
<linux/netdevice.h>.  Fixes these warnings:

../include/linux/netdevice.h:2100: warning: Function parameter or member 'gso_partial_features' not described in 'net_device'
../include/linux/netdevice.h:2100: warning: Function parameter or member 'l3mdev_ops' not described in 'net_device'
../include/linux/netdevice.h:2100: warning: Function parameter or member 'xfrmdev_ops' not described in 'net_device'
../include/linux/netdevice.h:2100: warning: Function parameter or member 'tlsdev_ops' not described in 'net_device'
../include/linux/netdevice.h:2100: warning: Function parameter or member 'name_assign_type' not described in 'net_device'
../include/linux/netdevice.h:2100: warning: Function parameter or member 'ieee802154_ptr' not described in 'net_device'
../include/linux/netdevice.h:2100: warning: Function parameter or member 'mpls_ptr' not described in 'net_device'
../include/linux/netdevice.h:2100: warning: Function parameter or member 'xdp_prog' not described in 'net_device'
../include/linux/netdevice.h:2100: warning: Function parameter or member 'gro_flush_timeout' not described in 'net_device'
../include/linux/netdevice.h:2100: warning: Function parameter or member 'xdp_bulkq' not described in 'net_device'
../include/linux/netdevice.h:2100: warning: Function parameter or member 'xps_cpus_map' not described in 'net_device'
../include/linux/netdevice.h:2100: warning: Function parameter or member 'xps_rxqs_map' not described in 'net_device'
../include/linux/netdevice.h:2100: warning: Function parameter or member 'qdisc_hash' not described in 'net_device'
../include/linux/netdevice.h:3552: WARNING: Inline emphasis start-string without end-string.
../include/linux/netdevice.h:3552: WARNING: Inline emphasis start-string without end-string.

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'dsa-headroom'
David S. Miller [Fri, 14 Feb 2020 15:34:51 +0000 (07:34 -0800)]
Merge branch 'dsa-headroom'

Per Forlin says:

====================
net: dsa: Make sure there is headroom for tag

Sorry for re-posting yet another time....
I manage to include multiple email-senders and forgot to include cover-letter.
Let's hope everyhthing is in order this time.

Fix two tag drivers to make sure there is headroom for the tag data.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: tag_ar9331: Make sure there is headroom for tag
Per Forlin [Thu, 13 Feb 2020 14:37:10 +0000 (15:37 +0100)]
net: dsa: tag_ar9331: Make sure there is headroom for tag

Passing tag size to skb_cow_head will make sure
there is enough headroom for the tag data.
This change does not introduce any overhead in case there
is already available headroom for tag.

Signed-off-by: Per Forlin <perfn@axis.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: tag_qca: Make sure there is headroom for tag
Per Forlin [Thu, 13 Feb 2020 14:37:09 +0000 (15:37 +0100)]
net: dsa: tag_qca: Make sure there is headroom for tag

Passing tag size to skb_cow_head will make sure
there is enough headroom for the tag data.
This change does not introduce any overhead in case there
is already available headroom for tag.

Signed-off-by: Per Forlin <perfn@axis.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet, ip6_tunnel: enhance tunnel locate with link check
William Dauchy [Thu, 13 Feb 2020 17:19:22 +0000 (18:19 +0100)]
net, ip6_tunnel: enhance tunnel locate with link check

With ipip, it is possible to create an extra interface explicitly
attached to a given physical interface:

  # ip link show tunl0
  4: tunl0@NONE: <NOARP> mtu 1480 qdisc noop state DOWN mode DEFAULT group default qlen 1000
    link/ipip 0.0.0.0 brd 0.0.0.0
  # ip link add tunl1 type ipip dev eth0
  # ip link show tunl1
  6: tunl1@eth0: <NOARP> mtu 1480 qdisc noop state DOWN mode DEFAULT group default qlen 1000
    link/ipip 0.0.0.0 brd 0.0.0.0

But it is not possible with ip6tnl:

  # ip link show ip6tnl0
  5: ip6tnl0@NONE: <NOARP> mtu 1452 qdisc noop state DOWN mode DEFAULT group default qlen 1000
      link/tunnel6 :: brd ::
  # ip link add ip6tnl1 type ip6tnl dev eth0
  RTNETLINK answers: File exists

This patch aims to make it possible by adding link comparaison in both
tunnel locate and lookup functions; we also modify mtu calculation when
attached to an interface with a lower mtu.

This permits to make use of x-netns communication by moving the newly
created tunnel in a given netns.

Signed-off-by: William Dauchy <w.dauchy@criteo.com>
Reviewed-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'mac80211-for-net-2020-02-14' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Fri, 14 Feb 2020 15:16:08 +0000 (07:16 -0800)]
Merge tag 'mac80211-for-net-2020-02-14' of git://git./linux/kernel/git/jberg/mac80211

Johannes Berg says:

====================
Just a few fixes:
 * avoid running out of tracking space for frames that need
   to be reported to userspace by using more bits
 * fix beacon handling suppression by adding some relevant
   elements to the CRC calculation
 * fix quiet mode in action frames
 * fix crash in ethtool for virt_wifi and similar
 * add a missing policy entry
 * fix 160 & 80+80 bandwidth to take local capabilities into
   account
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'smc-fixes'
David S. Miller [Fri, 14 Feb 2020 15:09:36 +0000 (07:09 -0800)]
Merge branch 'smc-fixes'

Karsten Graul says:

====================
net/smc: fixes for -net

Fix a syzbot finding and a problem with the CLC handshake content.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/smc: no peer ID in CLC decline for SMCD
Ursula Braun [Fri, 14 Feb 2020 07:59:00 +0000 (08:59 +0100)]
net/smc: no peer ID in CLC decline for SMCD

Just SMCR requires a CLC Peer ID, but not SMCD. The field should be
zero for SMCD.

Fixes: c758dfddc1b5 ("net/smc: add SMC-D support in CLC messages")
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/smc: transfer fasync_list in case of fallback
Ursula Braun [Fri, 14 Feb 2020 07:58:59 +0000 (08:58 +0100)]
net/smc: transfer fasync_list in case of fallback

SMC does not work together with FASTOPEN. If sendmsg() is called with
flag MSG_FASTOPEN in SMC_INIT state, the SMC-socket switches to
fallback mode. To handle the previous ioctl FIOASYNC call correctly
in this case, it is necessary to transfer the socket wait queue
fasync_list to the internal TCP socket.

Reported-by: syzbot+4b1fe8105f8044a26162@syzkaller.appspotmail.com
Fixes: ee9dfbef02d18 ("net/smc: handle sockopts forcing fallback")
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>