Config NPC layer info from KPU profile into protocol
checker to identify outer L2/IPv4/TCP/UDP headers in a
packet. And enable IPv4 checksum validation.

L3/L4 and L4 CSUM validation will be enabled by PF/VF
drivers by configuring NIX_AF_LF(0..127)_RX_CFG via mbox
i.e 'nix_lf_alloc_req->rx_cfg'

Also enable setting of NPC_RESULT_S[L2B] when an outer
L2 broadcast address is detected. This will help in
installing NPC MCAM rules for broadcast packets.
Signed-off-by: NSunil Goutham <sgoutham@marvell.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Added support for PF/VF drivers to configure NIX to
capture and/or strip VLAN tag from ingress packets.
Signed-off-by: NVamsi Attunuru <vamsi.attunuru@marvell.com>
Signed-off-by: NSunil Goutham <sgoutham@marvell.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Upon NIXLF ALLOC/FREE, add or remove corresponding PF_FUNC from
the broadcast packet replication list of the CGX LMAC mapped
RVU PF.
Signed-off-by: NSunil Goutham <sgoutham@marvell.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Allocate memory for mcast/bcast/mirror replication entry
contexts, replication buffers (used by HW) and config HW
with corresponding memory bases. Added support for installing
MCEs via NIX AQ mbox.

For now support is restricted to broadcast pkt replication,
hence MCE table size and number of replication buffers
allocated are less. Each CGX LMAC mapped RVU PF is assigned
a MCE table of size 'num VFs of that PF + PF'.
Signed-off-by: NSunil Goutham <sgoutham@marvell.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

For each CGX LMAC that is mapped to a RVU PF, allocate
a pkind and config the same in CGX. For a received packet
at CGX LMAC interface this pkind is used by NPC block
to start parsing of packet.
Signed-off-by: NGeetha sowjanya <gakula@marvell.com>
Signed-off-by: NSunil Goutham <sgoutham@marvell.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch configures all 16 KPUs and iKPU (pkinds) with
the KPU parser profile defined in npc_profile.h. Each KPU
engine has a 128 entry CAM, only CAM entries which are listed
in the profile are enabled and rest are left disabled.

Also
- Memory is allocated for pkind's bitmap and PFFUNC, interface
  channel mapping.
- Added all CSR offsets of NPC HW block.
Signed-off-by: NSunil Goutham <sgoutham@marvell.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

NPC block is responsible for parsing and forwarding
packets to different NIXLFs. NPC has 16 KPU engines
(Kangaroo parse engine) and one iKPU which represents
pkinds. Each physical port either CGX/LBK is assigned
a pkind and upon receiving a packet HW takes that port's
pkind and starts parsing as per the KPU engines config.

This patch adds header files which contain configuration
profile/array for each of the iKPU and 16 KPU engines.
Signed-off-by: NHao Zheng <hao.zheng@marvell.com>
Signed-off-by: NSunil Goutham <sgoutham@marvell.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds a new mailbox message to reset
a NIXLF's receive and transmit HW stats.
Signed-off-by: NVamsi Attunuru <vamsi.attunuru@marvell.com>
Signed-off-by: NSunil Goutham <sgoutham@marvell.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds support for a PF/VF driver to configure
NIX transmit scheduler queues via mbox. Since PF/VF doesn't
know the absolute HW index of the NIXLF attached to it, AF
traps the register config and overwrites with the correct
NIXLF index.

HW supports shaping, colouring and policing of packets with
these multilevel traffic scheduler queues. Instead of
introducing different mbox message formats for different
configurations and making both AF & PF/VF driver implementation
cumbersome, access to the scheduler queue's CSRs is provided
via mbox. AF checks whether the sender PF/VF has the
corresponding queue allocated or not and dumps the config
to HW. With a single mbox msg 20 registers can be configured.
Signed-off-by: NSunil Goutham <sgoutham@marvell.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Added support for a PF/VF to allocate or free NIX transmit
scheduler queues via mbox. For setting up pkt transmission
priorities between queues, the scheduler queues have to be
contiguous w.r.t their HW indices. So both contiguous and
non-contiguous allocations are supported.

Upon receiving NIX_TXSCH_FREE mbox msg all scheduler queues
allocated to sending PFFUNC (PF/VF) will be freed. Selective
free is not supported.
Signed-off-by: NSunil Goutham <sgoutham@marvell.com>
Signed-off-by: NNithin Dabilpuram <ndabilpuram@marvell.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

syzkaller triggered a use-after-free [1], caused by a combination of
skb_get() in llc_conn_state_process() and usage of sk_eat_skb()

sk_eat_skb() is assuming the skb about to be freed is only used by
the current thread. TCP/DCCP stacks enforce this because current
thread holds the socket lock.

llc_conn_state_process() wants to make sure skb does not disappear,
and holds a reference on the skb it manipulates. But as soon as this
skb is added to socket receive queue, another thread can consume it.

This means that llc must use regular skb_unlink() and kfree_skb()
so that both producer and consumer can safely work on the same skb.

[1]
BUG: KASAN: use-after-free in atomic_read include/asm-generic/atomic-instrumented.h:21 [inline]
BUG: KASAN: use-after-free in refcount_read include/linux/refcount.h:43 [inline]
BUG: KASAN: use-after-free in skb_unref include/linux/skbuff.h:967 [inline]
BUG: KASAN: use-after-free in kfree_skb+0xb7/0x580 net/core/skbuff.c:655
Read of size 4 at addr ffff8801d1f6fba4 by task ksoftirqd/1/18

CPU: 1 PID: 18 Comm: ksoftirqd/1 Not tainted 4.19.0-rc8+ #295
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+0x1c4/0x2b6 lib/dump_stack.c:113
 print_address_description.cold.8+0x9/0x1ff mm/kasan/report.c:256
 kasan_report_error mm/kasan/report.c:354 [inline]
 kasan_report.cold.9+0x242/0x309 mm/kasan/report.c:412
 check_memory_region_inline mm/kasan/kasan.c:260 [inline]
 check_memory_region+0x13e/0x1b0 mm/kasan/kasan.c:267
 kasan_check_read+0x11/0x20 mm/kasan/kasan.c:272
 atomic_read include/asm-generic/atomic-instrumented.h:21 [inline]
 refcount_read include/linux/refcount.h:43 [inline]
 skb_unref include/linux/skbuff.h:967 [inline]
 kfree_skb+0xb7/0x580 net/core/skbuff.c:655
 llc_sap_state_process+0x9b/0x550 net/llc/llc_sap.c:224
 llc_sap_rcv+0x156/0x1f0 net/llc/llc_sap.c:297
 llc_sap_handler+0x65e/0xf80 net/llc/llc_sap.c:438
 llc_rcv+0x79e/0xe20 net/llc/llc_input.c:208
 __netif_receive_skb_one_core+0x14d/0x200 net/core/dev.c:4913
 __netif_receive_skb+0x2c/0x1e0 net/core/dev.c:5023
 process_backlog+0x218/0x6f0 net/core/dev.c:5829
 napi_poll net/core/dev.c:6249 [inline]
 net_rx_action+0x7c5/0x1950 net/core/dev.c:6315
 __do_softirq+0x30c/0xb03 kernel/softirq.c:292
 run_ksoftirqd+0x94/0x100 kernel/softirq.c:653
 smpboot_thread_fn+0x68b/0xa00 kernel/smpboot.c:164
 kthread+0x35a/0x420 kernel/kthread.c:246
 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:413

Allocated by task 18:
 save_stack+0x43/0xd0 mm/kasan/kasan.c:448
 set_track mm/kasan/kasan.c:460 [inline]
 kasan_kmalloc+0xc7/0xe0 mm/kasan/kasan.c:553
 kasan_slab_alloc+0x12/0x20 mm/kasan/kasan.c:490
 kmem_cache_alloc_node+0x144/0x730 mm/slab.c:3644
 __alloc_skb+0x119/0x770 net/core/skbuff.c:193
 alloc_skb include/linux/skbuff.h:995 [inline]
 llc_alloc_frame+0xbc/0x370 net/llc/llc_sap.c:54
 llc_station_ac_send_xid_r net/llc/llc_station.c:52 [inline]
 llc_station_rcv+0x1dc/0x1420 net/llc/llc_station.c:111
 llc_rcv+0xc32/0xe20 net/llc/llc_input.c:220
 __netif_receive_skb_one_core+0x14d/0x200 net/core/dev.c:4913
 __netif_receive_skb+0x2c/0x1e0 net/core/dev.c:5023
 process_backlog+0x218/0x6f0 net/core/dev.c:5829
 napi_poll net/core/dev.c:6249 [inline]
 net_rx_action+0x7c5/0x1950 net/core/dev.c:6315
 __do_softirq+0x30c/0xb03 kernel/softirq.c:292

Freed by task 16383:
 save_stack+0x43/0xd0 mm/kasan/kasan.c:448
 set_track mm/kasan/kasan.c:460 [inline]
 __kasan_slab_free+0x102/0x150 mm/kasan/kasan.c:521
 kasan_slab_free+0xe/0x10 mm/kasan/kasan.c:528
 __cache_free mm/slab.c:3498 [inline]
 kmem_cache_free+0x83/0x290 mm/slab.c:3756
 kfree_skbmem+0x154/0x230 net/core/skbuff.c:582
 __kfree_skb+0x1d/0x20 net/core/skbuff.c:642
 sk_eat_skb include/net/sock.h:2366 [inline]
 llc_ui_recvmsg+0xec2/0x1610 net/llc/af_llc.c:882
 sock_recvmsg_nosec net/socket.c:794 [inline]
 sock_recvmsg+0xd0/0x110 net/socket.c:801
 ___sys_recvmsg+0x2b6/0x680 net/socket.c:2278
 __sys_recvmmsg+0x303/0xb90 net/socket.c:2390
 do_sys_recvmmsg+0x181/0x1a0 net/socket.c:2466
 __do_sys_recvmmsg net/socket.c:2484 [inline]
 __se_sys_recvmmsg net/socket.c:2480 [inline]
 __x64_sys_recvmmsg+0xbe/0x150 net/socket.c:2480
 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290
 entry_SYSCALL_64_after_hwframe+0x49/0xbe

The buggy address belongs to the object at ffff8801d1f6fac0
 which belongs to the cache skbuff_head_cache of size 232
The buggy address is located 228 bytes inside of
 232-byte region [ffff8801d1f6fac0, ffff8801d1f6fba8)
The buggy address belongs to the page:
page:ffffea000747dbc0 count:1 mapcount:0 mapping:ffff8801d9be7680 index:0xffff8801d1f6fe80
flags: 0x2fffc0000000100(slab)
raw: 02fffc0000000100 ffffea0007346e88 ffffea000705b108 ffff8801d9be7680
raw: ffff8801d1f6fe80 ffff8801d1f6f0c0 000000010000000b 0000000000000000
page dumped because: kasan: bad access detected

Memory state around the buggy address:
 ffff8801d1f6fa80: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb
 ffff8801d1f6fb00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
>ffff8801d1f6fb80: fb fb fb fb fb fc fc fc fc fc fc fc fc fc fc fc
                               ^
 ffff8801d1f6fc00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
 ffff8801d1f6fc80: fb fb fb fb fb fb fb fb fb fb fb fb fb fc fc fc

Fixes: 1da177e4 ("Linux-2.6.12-rc2")
Signed-off-by: NEric Dumazet <edumazet@google.com>
Reported-by: Nsyzbot <syzkaller@googlegroups.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Extract error information from rx and tx buffer descriptors,
and update error counters.
Signed-off-by: NMathias Thore <mathias.thore@infinera.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We should get 'driver_data' from 'struct device' directly. Going via
platform_device is an unneeded step back and forth.
Signed-off-by: NWolfram Sang <wsa+renesas@sang-engineering.com>
Reviewed-by: NAndrew Lunn <andrew@lunn.ch>
Reviewed-by: NFlorian Fainelli <f.fainelli@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We should get 'driver_data' from 'struct device' directly. Going via
platform_device is an unneeded step back and forth.
Signed-off-by: NWolfram Sang <wsa+renesas@sang-engineering.com>
Acked-by: NRichard Cochran <richardcochran@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

linux/prefetch.h is never explicitly included in ena_com, although
functions from it, such as prefetchw(), are used throughout ena_com.
This is an inclusion bug, and we fix it here by explicitly including
linux/prefetch.h. The bug was exposed when the driver was compiled
for the xtensa architecture.

Fixes: 689b2bda ("net: ena: add functions for handling Low Latency Queues in ena_com")
Fixes: 8c590f97 ("ena: Fix Kconfig dependency on X86")
Signed-off-by: NArthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Replace spurious spaces with a tab and remove superfluous tab from
unix_sock struct.
Signed-off-by: NVito Caputo <vcaputo@pengaru.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

FDB_MAX_CHAIN is three.  We wanted to allocate enough memory to hold four
structs but there are missing parentheses so we only allocate enough
memory for three structs and the first byte of the fourth one.

Fixes: 328edb49 ("net/mlx5: Split FDB fast path prio to multiple namespaces")
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Davide Caratti says:

====================
net/sched: forbid 'goto_chain' on fallback actions

the following command:

 # tc actions add action police rate 1mbit burst 1k conform-exceed \
 > pass / goto chain 42

generates a NULL pointer dereference when packets exceed the configured
rate. Similarly, the following command:

 # tc actions add action pass random determ goto chain 42 2

makes the kernel crash with NULL dereference when the first packet does
not match the 'pass' action.

gact and police allow users to specify a fallback control action, that is
stored in the action private data. 'goto chain x' never worked for these
cases, since a->goto_chain handle was never initialized. There is only one
goto_chain handle per TC action, and it is designed to be non-NULL only if
tcf_action contains a 'goto chain' command. So, let's forbid 'goto chain'
on fallback actions.

Patch 1/4 and 2/4 change the .init() functions of police and gact, to let
them return an error when users try to set 'goto chain x' in the fallback
action. Patch 3/4 and 4/4 add TDC selftest coverage to this new behavior.
====================
Acked-by: NJamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

add test to verify if act_police forbids 'goto chain' control actions for
'exceed' traffic.
Signed-off-by: NDavide Caratti <dcaratti@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

add test to verify if act_gact forbids 'goto chain' control actions on
'random' traffic in gact.json.
Signed-off-by: NDavide Caratti <dcaratti@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

in the following command:

 # tc action add action police rate <r> burst <b> conform-exceed <c1>/<c2>

'goto chain x' is allowed only for c1: setting it for c2 makes the kernel
crash with NULL pointer dereference, since TC core doesn't initialize the
chain handle.
Signed-off-by: NDavide Caratti <dcaratti@redhat.com>
Acked-by: NCong Wang <xiyou.wangcong@gmail.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

in the following command:

 # tc action add action <c1> random <rand_type> <c2> <rand_param>

'goto chain x' is allowed only for c1: setting it for c2 makes the kernel
crash with NULL pointer dereference, since TC core doesn't initialize the
chain handle.
Signed-off-by: NDavide Caratti <dcaratti@redhat.com>
Acked-by: NCong Wang <xiyou.wangcong@gmail.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When indicating the MAC supports Symmetric Pause, clear the Asymmetric
Pause bit, which could of been already set is the PHY supports it.
Reported-by: NLabbe Corentin <clabbe@baylibre.com>
Fixes: c306ad36 ("net: ethernet: Add helper for MACs which support pause")
Signed-off-by: NAndrew Lunn <andrew@lunn.ch>
Tested-by: NCorentin Labbe <clabbe@baylibre.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NOlivier Brunel <jjk@jjacky.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

User mode helpers were spawned without a command line, and because
an empty command line is used by many tools to identify processes as
kernel threads, this could cause some issues.

Notably during killing spree on shutdown, since such helper would then
be skipped (i.e. not killed) which would result in the process remaining
alive, and thus preventing unmouting of the rootfs (as experienced with
the bpfilter umh).

Fixes: 449325b5 ("umh: introduce fork_usermode_blob() helper")
Signed-off-by: NOlivier Brunel <jjk@jjacky.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

These functions are supposed to return one on failure and zero on
success.  Returning a zero here could cause uninitialized variable
bugs in several of the callers.  For example:

    drivers/scsi/cxgbi/cxgb4i/cxgb4i.c:1660 get_iscsi_dcb_priority()
    error: uninitialized symbol 'caps'.

Fixes: 48365e48 ("qlcnic: dcb: Add support for CEE Netlink interface.")
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

David Ahern says:

====================
net: Add support for dumping addresses for a specific device

Use the recently added kernel side filter infrastructure to add support
for dumping addresses only for a specific device.

Patch 1 creates an IPv4 version similar to IPv6's in6_dump_addrs function.

Patch 2 simplifies in6_dump_addrs by moving index tracking of IP
addresses from inet6_dump_addr to in6_dump_addrs.

Patches 3 and 4 use the device-based address dump helpers to limit a
dump to just the addresses on a specific device.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If an RTM_GETADDR dump request has ifa_index set in the ifaddrmsg
header, then return only the addresses for that device.

Since inet6_dump_addr is reused for multicast and anycast addresses,
this adds support for device specfic dumps of RTM_GETMULTICAST and
RTM_GETANYCAST as well.
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If an RTM_GETADDR dump request has ifa_index set in the ifaddrmsg
header, then return only the addresses for that device.
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

ip_idx is always 0 going into in6_dump_addrs; it is passed as a pointer
to save the last good index into cb. Since cb is already argument to
in6_dump_addrs, just save the value there.
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Similar to IPv6 move the logic that walks over the ipv4 address list
for a device into a helper.
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Salil Mehta says:

====================
Adds support of RAS Error Handling in HNS3 Driver

This patch-set adds support related to RAS Error handling to the HNS3
Ethernet PF Driver. Set of errors occurred in the HNS3 hardware are
reported to the driver through the PCIe AER interface. The received
error information is then used to classify the received errors and
then decide the appropriate receovery action depending on the type
of error.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch enables and process hw errors of TM scheduler and
QCN(Quantized Congestion Control).
Signed-off-by: NShiju Jose <shiju.jose@huawei.com>
Signed-off-by: NSalil Mehta <salil.mehta@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch enables and process hw errors from the
PPP(Programmable Packet Process) block.
Signed-off-by: NShiju Jose <shiju.jose@huawei.com>
Signed-off-by: NSalil Mehta <salil.mehta@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds enable and processing of hw errors from IGU(Ingress Unit),
EGU(Egress Unit) and NCSI(Network Controller Sideband Interface).
Signed-off-by: NShiju Jose <shiju.jose@huawei.com>
Signed-off-by: NSalil Mehta <salil.mehta@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds enable and processing of ecc errors from
common HNS blocks, CMDQ(Command Queue),
IMP(Integrated Management Processor) and TQP(Task Queue Pair).
Signed-off-by: NShiju Jose <shiju.jose@huawei.com>
Signed-off-by: NSalil Mehta <salil.mehta@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds functions to enable and disable hw errors.
Signed-off-by: NShiju Jose <shiju.jose@huawei.com>
Signed-off-by: NSalil Mehta <salil.mehta@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds the error recovery for the HNS hw errors.
Signed-off-by: NShiju Jose <shiju.jose@huawei.com>
Signed-off-by: NSalil Mehta <salil.mehta@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Set of hw errors occurred in the HNS3 are reported to the
hns3 driver through PCIe AER and RAS.The error info will be
processed and appropriately recovered.
This patch adds error_detected callback and error processing.
Signed-off-by: NShiju Jose <shiju.jose@huawei.com>
Signed-off-by: NSalil Mehta <salil.mehta@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Clang warns (trimmed for brevity):

drivers/isdn/mISDN/tei.c:1193:7: warning: overflow converting case value
to switch condition type (2147764552 to 18446744071562348872) [-Wswitch]
        case IMHOLD_L1:
             ^
drivers/isdn/mISDN/tei.c:1187:7: warning: overflow converting case value
to switch condition type (2147764550 to 18446744071562348870) [-Wswitch]
        case IMCLEAR_L2:
             ^
2 warnings generated.

The root cause is that the _IOC macro can generate really large numbers,
which don't find into type int. My research into how GCC and Clang are
handling this at a low level didn't prove fruitful and surveying the
kernel tree shows that aside from here and a few places in the scsi
subsystem, everything that uses _IOC is at least of type 'unsigned int'.
Make that change here because as nothing in this function cares about
the signedness of the variable and it removes ambiguity, which is never
good when dealing with compilers.

While we're here, remove the unnecessary local variable ret (just return
-EINVAL and 0 directly).

Link: https://github.com/ClangBuiltLinux/linux/issues/67Signed-off-by: NNathan Chancellor <natechancellor@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>