syzkaller had no problem to trigger a deadlock, attaching a KCM socket
to another one (or itself). (original syzkaller report was a very
confusing lockdep splat during a sendmsg())
It seems KCM claims to only support TCP, but no enforcement is done,
so we might need to add additional checks.
Fixes: ab7ac4eb98 ("kcm: Kernel Connection Multiplexor module")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Acked-by: Tom Herbert <tom@quantonium.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Nikolay Aleksandrov says:
====================
net/sched: init failure fixes
I went over all qdiscs' init, destroy and reset callbacks and found the
issues fixed in each patch. Mostly they are null pointer dereferences due
to uninitialized timer (qdisc watchdog) or double frees due to ->destroy
cleaning up a second time. There's more information in each patch.
I've tested these by either sending wrong attributes from user-spaces, no
attributes or by simulating memory alloc failure where applicable. Also
tried all of the qdiscs as a default qdisc.
Most of these bugs were present before commit 87b60cfacf, I've tried to
include proper fixes tags in each patch.
I haven't included individual patch acks in the set, I'd appreciate it if
you take another look and resend them.
====================
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently only a memory allocation failure can lead to this, so let's
initialize the timer first.
Fixes: 6529eaba33 ("net: sched: introduce tcf block infractructure")
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is very unlikely to happen but the backlogs memory allocation
could fail and will free q->flows, but then ->destroy() will free
q->flows too. For correctness remove the first free and let ->destroy
clean up.
Fixes: 87b60cfacf ("net_sched: fix error recovery at qdisc creation")
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Depending on where ->init fails we can get a null pointer deref due to
uninitialized hires timer (watchdog) or a double free of the qdisc hash
because it is already freed by ->destroy().
Fixes: 8d55373875 ("net/sched/hfsc: allocate tcf block for hfsc root class")
Fixes: 87b60cfacf ("net_sched: fix error recovery at qdisc creation")
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is quite common for ti_cm_get_macid() to fail on some of the
platforms it is invoked on. They include any platform where
mac address is not part of SoC register space.
On these platforms, mac address is read and populated in
device-tree by bootloader. An example is TI DA850.
Downgrade the severity of message to "information", so it does
not spam logs when 'quiet' boot is desired.
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The phy is connected at early stage of probe but not properly
disconnected if error occurs. This patch fixes the issue.
Also changing the return type of xgene_enet_check_phy_handle(),
since this function always returns success.
Signed-off-by: Quan Nguyen <qnguyen@apm.com>
Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Both the nfp_net_pf_app_start() and the nfp_net_pci_probe() functions
call nfp_net_pf_app_stop_ctrl(pf) so there is a double free. The free
should be done from the probe function because it's allocated there so
I have removed the call from nfp_net_pf_app_start().
Fixes: 02082701b9 ("nfp: create control vNICs and wire up rx/tx")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pavel Belous says:
====================
net:ethernet:aquantia: Atlantic driver Update 2017-08-23
This series contains updates for aQuantia Atlantic driver.
It has bugfixes and some improvements.
Changes in v2:
- "MCP state change" fix removed (will be sent as
a separate fix after further investigation.)
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
We should inform user about wrong firmware version
by printing message in dmesg.
Fixes: 3d2ff7eebe26 ("net: ethernet: aquantia: Atlantic hardware abstraction layer")
Signed-off-by: Pavel Belous <Pavel.Belous@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since the HW supports up to 32 multicast filters we should
track count of multicast filters to avoid overflow.
If we attempt to add >32 multicast filter - just set NETIF_ALLMULTI flag
instead.
Fixes: 94f6c9e4cdf6 ("net: ethernet: aquantia: Support for NIC-specific code")
Signed-off-by: Igor Russkikh <Igor.Russkikh@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver choose the optimal interrupt throttling settings depends
of current link speed.
Due this bug link_status field from aq_hw is never updated and as result
always used same interrupt throttling values.
Fixes: 3d2ff7eebe26 ("net: ethernet: aquantia: Atlantic hardware abstraction layer")
Signed-off-by: Pavel Belous <Pavel.Belous@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The hardware has the HW Checksum Offload bug when small
TCP patckets (with length <= 60 bytes) has wrong "checksum valid" bit.
The solution is - ignore checksum valid bit for small packets
(with length <= 60 bytes) and mark this as CHECKSUM_NONE to allow
network stack recalculate checksum itself.
Fixes: ccf9a5ed14be ("net: ethernet: aquantia: Atlantic A0 and B0 specific functions.")
Signed-off-by: Pavel Belous <Pavel.Belous@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The number of RSS queues should be not more than numbers of CPU.
Its does not make sense to increase perfomance, and also cause problems on
some motherboards.
Fixes: 94f6c9e4cdf6 ("net: ethernet: aquantia: Support for NIC-specific code")
Signed-off-by: Pavel Belous <Pavel.Belous@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes datapath spinlocks which does not perform any
useful work.
Fixes: 6e70637f9f1e ("net: ethernet: aquantia: Add ring support code")
Signed-off-by: Pavel Belous <Pavel.Belous@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
... which may happen with certain values of tp_reserve and maclen.
Fixes: 58d19b19cd ("packet: vnet_hdr support for tpacket_rcv")
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Cc: Willem de Bruijn <willemb@google.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For a bond slave device as a tipc bearer, the dev represents the bond
interface and orig_dev represents the slave in tipc_l2_rcv_msg().
Since we decode the tipc_ptr from bonding device (dev), we fail to
find the bearer and thus tipc links are not established.
In this commit, we register the tipc protocol callback per device and
look for tipc bearer from both the devices.
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ChunYu found a kernel warn_on during syzkaller fuzzing:
[40226.038539] WARNING: CPU: 5 PID: 23720 at net/ipv4/af_inet.c:152 inet_sock_destruct+0x78d/0x9a0
[40226.144849] Call Trace:
[40226.147590] <IRQ>
[40226.149859] dump_stack+0xe2/0x186
[40226.176546] __warn+0x1a4/0x1e0
[40226.180066] warn_slowpath_null+0x31/0x40
[40226.184555] inet_sock_destruct+0x78d/0x9a0
[40226.246355] __sk_destruct+0xfa/0x8c0
[40226.290612] rcu_process_callbacks+0xaa0/0x18a0
[40226.336816] __do_softirq+0x241/0x75e
[40226.367758] irq_exit+0x1f6/0x220
[40226.371458] smp_apic_timer_interrupt+0x7b/0xa0
[40226.376507] apic_timer_interrupt+0x93/0xa0
The warn_on happned when sk->sk_rmem_alloc wasn't 0 in inet_sock_destruct.
As after commit f970bd9e3a ("udp: implement memory accounting helpers"),
udp has changed to use udp_destruct_sock as sk_destruct where it would
udp_rmem_release all rmem.
But IPV6_ADDRFORM sockopt sets sk_destruct with inet_sock_destruct after
changing family to PF_INET. If rmem is not 0 at that time, and there is
no place to release rmem before calling inet_sock_destruct, the warn_on
will be triggered.
This patch is to fix it by not setting sk_destruct in IPV6_ADDRFORM sockopt
any more. As IPV6_ADDRFORM sockopt only works for tcp and udp. TCP sock has
already set it's sk_destruct with inet_sock_destruct and UDP has set with
udp_destruct_sock since they're created.
Fixes: f970bd9e3a ("udp: implement memory accounting helpers")
Reported-by: ChunYu Wang <chunwang@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Steffen Klassert says:
====================
pull request (net): ipsec 2017-08-29
1) Fix dst_entry refcount imbalance when using socket policies.
From Lorenzo Colitti.
2) Fix locking when adding the ESP trailers.
3) Fix tailroom calculation for the ESP trailer by using
skb_tailroom instead of skb_availroom.
4) Fix some info leaks in xfrm_user.
From Mathias Krause.
Please pull or let me know if there are problems.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Now it doesn't check for the cached route expiration in ipv6's
dst_ops->check(), because it trusts dst_gc that would clean the
cached route up when it's expired.
The problem is in dst_gc, it would clean the cached route only
when it's refcount is 1. If some other module (like xfrm) keeps
holding it and the module only release it when dst_ops->check()
fails.
But without checking for the cached route expiration, .check()
may always return true. Meanwhile, without releasing the cached
route, dst_gc couldn't del it. It will cause this cached route
never to expire.
This patch is to set dst.obsolete with DST_OBSOLETE_KILL in .gc
when it's expired, and check obsolete != DST_OBSOLETE_FORCE_CHK
in .check.
Note that this is even needed when ipv6 dst_gc timer is removed
one day. It would set dst.obsolete in .redirect and .update_pmtu
instead, and check for cached route expiration when getting it,
just like what ipv4 route does.
Reported-by: Jianlin Shi <jishi@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit c5cff8561d adds rcu grace period before freeing fib6_node. This
generates a new sparse warning on rt->rt6i_node related code:
net/ipv6/route.c:1394:30: error: incompatible types in comparison
expression (different address spaces)
./include/net/ip6_fib.h:187:14: error: incompatible types in comparison
expression (different address spaces)
This commit adds "__rcu" tag for rt6i_node and makes sure corresponding
rcu API is used for it.
After this fix, sparse no longer generates the above warning.
Fixes: c5cff8561d ("ipv6: add rcu grace period before freeing fib6_node")
Signed-off-by: Wei Wang <weiwan@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Passing commands for logging to t4_record_mbox() with size
MBOX_LEN, when the actual command size is actually smaller,
causes out-of-bounds stack accesses in t4_record_mbox() while
copying command words here:
for (i = 0; i < size / 8; i++)
entry->cmd[i] = be64_to_cpu(cmd[i]);
Up to 48 bytes from the stack are then leaked to debugfs.
This happens whenever we send (and log) commands described by
structs fw_sched_cmd (32 bytes leaked), fw_vi_rxmode_cmd (48),
fw_hello_cmd (48), fw_bye_cmd (48), fw_initialize_cmd (48),
fw_reset_cmd (48), fw_pfvf_cmd (32), fw_eq_eth_cmd (16),
fw_eq_ctrl_cmd (32), fw_eq_ofld_cmd (32), fw_acl_mac_cmd(16),
fw_rss_glb_config_cmd(32), fw_rss_vi_config_cmd(32),
fw_devlog_cmd(32), fw_vi_enable_cmd(48), fw_port_cmd(32),
fw_sched_cmd(32), fw_devlog_cmd(32).
The cxgb4vf driver got this right instead.
When we call t4_record_mbox() to log a command reply, a MBOX_LEN
size can be used though, as get_mbox_rpl() will fill cmd_rpl up
completely.
Fixes: 7f080c3f2f ("cxgb4: Add support to enable logging of firmware mailbox commands")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since the bindings have been controversial, and we follow the DT stable ABI
rule, we shouldn't let a driver with a DT binding that might change slip
through in a stable release.
Remove the compatibles to make sure the driver will not probe and no-one
will start using the binding currently implemented. This commit will
obviously need to be reverted in due time.
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pieter Jansen van Vuuren says:
====================
nfp: fix layer calculation and flow dissector use
Previously when calculating the supported key layers MPLS, IPv4/6
TTL and TOS were not considered. Formerly flow dissectors were referenced
without first checking that they are in use and correctly populated by TC.
Additionally this patch set fixes the incorrect use of mask field for vlan
matching.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Previously the vlan tci field was incorrectly exact matched. This patch
fixes this by using the flow dissector to populate the vlan tci field.
Fixes: 5571e8c9f2 ("nfp: extend flower matching capabilities")
Signed-off-by: Pieter Jansen van Vuuren <pieter.jansenvanvuuren@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Previously when calculating the supported key layers MPLS, IPv4/6
TTL and TOS were not considered. This patch checks that the TTL and
TOS fields are masked out before offloading. Additionally this patch
checks that MPLS packets are correctly handled, by not offloading them.
Fixes: af9d842c13 ("nfp: extend flower add flow offload")
Signed-off-by: Pieter Jansen van Vuuren <pieter.jansenvanvuuren@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Previously flow dissectors were referenced without first checking that
they are in use and correctly populated by TC. This patch fixes this by
checking each flow dissector key before referencing them.
Fixes: 5571e8c9f2 ("nfp: extend flower matching capabilities")
Signed-off-by: Pieter Jansen van Vuuren <pieter.jansenvanvuuren@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Guillaume Nault says:
====================
l2tp: fix some l2tp_tunnel_find() issues in l2tp_netlink
Since l2tp_tunnel_find() doesn't take a reference on the tunnel it
returns, its users are almost guaranteed to be racy.
This series defines l2tp_tunnel_get() which can be used as a safe
replacement, and converts some of l2tp_tunnel_find() users in the
l2tp_netlink module.
Other users often combine this issue with other more or less subtle
races. They will be fixed incrementally in followup series.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Use l2tp_tunnel_get() to retrieve tunnel, so that it can't go away on
us. Otherwise l2tp_tunnel_destruct() might release the last reference
count concurrently, thus freeing the tunnel while we're using it.
Fixes: 309795f4be ("l2tp: Add netlink control API for L2TP")
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use l2tp_tunnel_get() instead of l2tp_tunnel_find() so that we get
a reference on the tunnel, preventing l2tp_tunnel_destruct() from
freeing it from under us.
Also move l2tp_tunnel_get() below nlmsg_new() so that we only take
the reference when needed.
Fixes: 309795f4be ("l2tp: Add netlink control API for L2TP")
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
We need to make sure the tunnel is not going to be destroyed by
l2tp_tunnel_destruct() concurrently.
Fixes: 309795f4be ("l2tp: Add netlink control API for L2TP")
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
l2tp_nl_cmd_tunnel_delete() needs to take a reference on the tunnel, to
prevent it from being concurrently freed by l2tp_tunnel_destruct().
Fixes: 309795f4be ("l2tp: Add netlink control API for L2TP")
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
l2tp_tunnel_find() doesn't take a reference on the returned tunnel.
Therefore, it's unsafe to use it because the returned tunnel can go
away on us anytime.
Fix this by defining l2tp_tunnel_get(), which works like
l2tp_tunnel_find(), but takes a reference on the returned tunnel.
Caller then has to drop this reference using l2tp_tunnel_dec_refcount().
As l2tp_tunnel_dec_refcount() needs to be moved to l2tp_core.h, let's
simplify the patch and not move the L2TP_REFCNT_DEBUG part. This code
has been broken (not even compiling) in May 2012 by
commit a4ca44fa57 ("net: l2tp: Standardize logging styles")
and fixed more than two years later by
commit 29abe2fda5 ("l2tp: fix missing line continuation"). So it
doesn't appear to be used by anyone.
Same thing for l2tp_tunnel_free(); instead of moving it to l2tp_core.h,
let's just simplify things and call kfree_rcu() directly in
l2tp_tunnel_dec_refcount(). Extra assertions and debugging code
provided by l2tp_tunnel_free() didn't help catching any of the
reference counting and socket handling issues found while working on
this series.
Fixes: 309795f4be ("l2tp: Add netlink control API for L2TP")
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
Sessions must be fully initialised before calling
l2tp_session_add_to_tunnel(). Otherwise, there's a short time frame
where partially initialised sessions can be accessed by external users.
Fixes: dbdbc73b44 ("l2tp: fix duplicate session creation")
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
The mac address is only retrieved from h/w when using PPv2.1. Otherwise
the variable holding it is still checked and used if it contains a valid
value. As the variable isn't initialized to an invalid mac address
value, we end up with random mac addresses which can be the same for all
the ports handled by this PPv2 driver.
Fixes this by initializing the h/w mac address variable to {0}, which is
an invalid mac address value. This way the random assignation fallback
is called and all ports end up with their own addresses.
Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Fixes: 2697582144 ("net: mvpp2: handle misc PPv2.1/PPv2.2 differences")
Signed-off-by: David S. Miller <davem@davemloft.net>
The u-blox TOBY-L4 is a LTE Advanced (Cat 6) module with HSPA+ and 2G
fallback.
Unlike the TOBY-L2, this module has one single USB layout and exposes
several TTYs for control and a NCM interface for data. Connecting this
module may be done just by activating the desired PDP context with
'AT+CGACT=1,<cid>' and then running DHCP on the NCM interface.
Signed-off-by: Aleksander Morgado <aleksander@aleksander.es>
Signed-off-by: David S. Miller <davem@davemloft.net>
Noticed that busy_poll_stop() also invoke the drivers napi->poll()
function pointer, but didn't have an associated call to trace_napi_poll()
like all other call sites.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The memory reserved to dump the ID of the xfrm state includes a padding
byte in struct xfrm_usersa_id added by the compiler for alignment. To
prevent the heap info leak, memset(0) the sa_id before filling it.
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Fixes: d51d081d65 ("[IPSEC]: Sync series - user")
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
The memory reserved to dump the expired xfrm state includes padding
bytes in struct xfrm_user_expire added by the compiler for alignment. To
prevent the heap info leak, memset(0) the remainder of the struct.
Initializing the whole structure isn't needed as copy_to_user_state()
already takes care of clearing the padding bytes within the 'state'
member.
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
The memory reserved to dump the ID of the xfrm state includes a padding
byte in struct xfrm_usersa_id added by the compiler for alignment. To
prevent the heap info leak, memset(0) the whole struct before filling
it.
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Fixes: 0603eac0d6 ("[IPSEC]: Add XFRMA_SA/XFRMA_POLICY for delete notification")
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
The memory reserved to dump the xfrm offload state includes padding
bytes of struct xfrm_user_offload added by the compiler for alignment.
Add an explicit memset(0) before filling the buffer to avoid the heap
info leak.
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Fixes: d77e38e612 ("xfrm: Add an IPsec hardware offloading API")
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Only one iwlwifi patch this time.
iwlwifi
* fix multiple times reported lockdep warning found by new locking
annotation introduced in v4.13-rc1
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJZoCT5AAoJEG4XJFUm622b1Y8H/iiQbFBDv9uyoGYZOjbJtpfY
B+hJoytPOGsuCQSjDApM7/nfjJcHOwESpN0CFaHwusumMFiBIMVKBTrml8RI7jmE
NBUnsBld1fJzsB1ErlI0vHN7YXym38VD280q15jEVa+XaGEo7l51nJGGLfmw64Ac
EPvBoLsTsPbpsNpzfHLCYuUcucTEREi90P+qek4uzmikcf07bIa14h0WC0XG6QdQ
YJpRNhWH4znvqwEltzNtpADzRiQma3pXu9JwtgTfaPy1vFbVaSzaDNqLdt8UE6IL
ANVAXBQQFR887f0NKwAVFfYDD50XT2ThBvsCEVQ4fecf8r3x1WfoSIEvvwGcmuQ=
=Sn3N
-----END PGP SIGNATURE-----
Merge tag 'wireless-drivers-for-davem-2017-08-25' of git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers
Kalle Valo says:
====================
wireless-drivers fixes for 4.13
Only one iwlwifi patch this time.
iwlwifi
* fix multiple times reported lockdep warning found by new locking
annotation introduced in v4.13-rc1
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
