firefly-linux-kernel-4.4.55.git
11 years agoigb: Support using build_skb in the case that jumbo frames are disabled
Alexander Duyck [Sat, 2 Feb 2013 05:07:11 +0000 (05:07 +0000)]
igb: Support using build_skb in the case that jumbo frames are disabled

This change makes it so that we can enable the use of build_skb for cases
where jumbo frames are disabled.  The advantage to this is that we do not
have to perform a memcpy to populate the header and as a result we see a
significant performance improvement.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
11 years agonet: Don't write to current task flags on every packet received.
David S. Miller [Thu, 14 Feb 2013 20:57:38 +0000 (15:57 -0500)]
net: Don't write to current task flags on every packet received.

Even for non-pfmalloc SKBs, __netif_receive_skb() will do a
tsk_restore_flags() on current unconditionally.

Make __netif_receive_skb() a shim around the existing code, renamed to
__netif_receive_skb_core().  Let __netif_receive_skb() wrap the
__netif_receive_skb_core() call with the task flag modifications, if
necessary.

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agogianfar: Fix and cleanup Rx FCB indication
Claudiu Manoil [Thu, 14 Feb 2013 05:00:07 +0000 (05:00 +0000)]
gianfar: Fix and cleanup Rx FCB indication

This fixes a less obvious error on one hand, and prevents futher
similar errors by disambiguating and optimizing RxFCB indication,
on the other hand.

The error consists in NETIF_F_HW_VLAN_TX flag being used as an
indication of Rx FCB insertion. This happened as soon gfar_uses_fcb(),
which despite its name indicates Rx FCB insertion, started
incorporating is_vlan_on().
is_vlan_on(), on the other hand, is also a misleading construct because
we need to differentiate b/w hw VLAN extraction/VLEX (marked by VLAN_RX
flag) and hw VLAN insertion/VLINS (VLAN_TX flag), which are different
mechanisms using different types of FCBs.

The hw spec for the RxFCB feature is as follows:
In the case of RxBD rings, FCBs (Frame Control Block) are inserted by
the eTSEC whenever RCTRL[PRSDEP] is set to a non-zero value. Only one
FCB is inserted per frame (in the buffer pointed to by the RxBD with
bit F set). TOE acceleration for receive is enabled for all rx frames
in this case.

This patch introduces priv->uses_rxfcb field to quickly signal RxFCB
insertion in accordance with the specification above.

The dependency on FSL_GIANFAR_DEV_HAS_TIMER was also eliminated as
another source of confusion. The actual dependency is to priv->hwts_rx_en.
Upon changing priv->hwts_rx_en via IOCTL, the gfar device is being
restarted and on init_mac() the priv->hwts_rx_en flag determines RxFCB
insertion, and rctrl is programmed accordingly. The patch takes care
of this case too.

Though maybe not as self documenting as the inlining version uses_fcb(),
priv->uses_rxfcb has the main purpose to quickly signal, on the hot path,
that the incoming frame has a *Rx* FCB block inserted which needs to be
pulled out before passing the skb to the stack. This is a performance
critical operation, it needs to happen fast, that's why uses_rxfcb is
placed in the first cacheline of gfar_private.
This is also why a cached rctrl cannot be used instead: 1) because
we don't have 32 bits available in the first cacheline of gfar_priv
(but only 16); 2) bit operations are expensive on the hot path.

Signed-off-by: Claudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agogianfar: Remove wrong buffer size conditioning to VLAN h/w offload
Claudiu Manoil [Thu, 14 Feb 2013 05:00:06 +0000 (05:00 +0000)]
gianfar: Remove wrong buffer size conditioning to VLAN h/w offload

The controller's ref manual states clearly that when the hw Rx vlan
offload feature is enabled, meaning that the VLEX bit from RCTRL is
correctly enabled, then the hw performs automatic VLAN tag extraction
and deletion from the ethernet frames. So there's no point in trying to
increase the rx buff size when rxvlan is on, as the frame is actually
smaller.
And the Tx vlan hw accel feature (VLINS) has nothing to do with rx buff
size computation.

Signed-off-by: Claudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agogianfar: gfar_process_frame returns void
Claudiu Manoil [Thu, 14 Feb 2013 05:00:05 +0000 (05:00 +0000)]
gianfar: gfar_process_frame returns void

No return code is expected from gfar_process_frame(), hence
change it to return void.

Signed-off-by: Claudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agogianfar: GRO_DROP is unlikely
Claudiu Manoil [Thu, 14 Feb 2013 05:00:04 +0000 (05:00 +0000)]
gianfar: GRO_DROP is unlikely

The change is significant since it affects the rx hot path.
Paul observed and documented the effects at asm level, see
below:

"It turns out that it does make a difference, since gfar_process_frame
gets inlined, and so the increment code gets moved out of line (I have
marked the if statment with * and the increment code within "-----"):

  ------------------------- as is currently ------------------
     4d14:       80 61 00 18     lwz     r3,24(r1)
     4d18:       7f c4 f3 78     mr      r4,r30
     4d1c:       48 00 00 01     bl      4d1c <gfar_clean_rx_ring+0x10c>
  *  4d20:       2f 83 00 04     cmpwi   cr7,r3,4
     4d24:       40 9e 00 1c     bne-    cr7,4d40
<gfar_clean_rx_ring+0x130>
        ----------------------------
     4d28:       81 3c 01 f8     lwz     r9,504(r28)
     4d2c:       81 5c 01 fc     lwz     r10,508(r28)
     4d30:       31 4a 00 01     addic   r10,r10,1
     4d34:       7d 29 01 94     addze   r9,r9
     4d38:       91 3c 01 f8     stw     r9,504(r28)
     4d3c:       91 5c 01 fc     stw     r10,508(r28)
        ----------------------------
     4d40:       a0 1f 00 24     lhz     r0,36(r31)
     4d44:       81 3f 00 00     lwz     r9,0(r31)
     4d48:       7f a4 eb 78     mr      r4,r29
     4d4c:       7f e3 fb 78     mr      r3,r31

  -------------------------- unlikely ------------------------
     4d14:       80 61 00 18     lwz     r3,24(r1)
     4d18:       7f c4 f3 78     mr      r4,r30
     4d1c:       48 00 00 01     bl      4d1c <gfar_clean_rx_ring+0x10c>
  *  4d20:       2f 83 00 04     cmpwi   cr7,r3,4
     4d24:       41 9e 03 94     beq-    cr7,50b8
<gfar_clean_rx_ring+0x4a8>
     4d28:       a0 1f 00 24     lhz     r0,36(r31)
     4d2c:       81 3f 00 00     lwz     r9,0(r31)
     4d30:       7f a4 eb 78     mr      r4,r29
     4d34:       7f e3 fb 78     mr      r3,r31
[...]
     50b8:       81 3c 01 f8     lwz     r9,504(r28)
     50bc:       81 5c 01 fc     lwz     r10,508(r28)
     50c0:       31 4a 00 01     addic   r10,r10,1
     50c4:       7d 29 01 94     addze   r9,r9
     50c8:       91 3c 01 f8     stw     r9,504(r28)
     50cc:       91 5c 01 fc     stw     r10,508(r28)
     50d0:       4b ff fc 58     b       4d28 <gfar_clean_rx_ring+0x118>

So, the increment does actually get moved ~1k away."

Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Claudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agogianfar: Cleanup and optimize struct gfar_private
Claudiu Manoil [Thu, 14 Feb 2013 05:00:03 +0000 (05:00 +0000)]
gianfar: Cleanup and optimize struct gfar_private

Group run-time critical fields within the 1st cacheline (32B)
followed by the tx|rx_queue reference arrays and the interrupt
group instances (gfargrp), all cacheline aligned.

This has several benefits. Firstly comes the performance benefit
by having the members required by the driver's hot path re-grouped
in the structure's first cache lines, whereas the unimportant
members were pushed towards the end of the struct.
Another benefit comes from eliminating a 24 byte memory hole that
was rendering gfar_priv's 2nd cacheline useless. The default gcc
layout of gfar_private leaves an implicit 24 byte hole after the
errata (enum) member. This patch fixes it.

The uchar bitfields were pushed towards the end of the struct
as these are not run-time performance critical (used for init
time operations). Because there is no other 2 byte member
around to couple the uchar bitfields memeber with, we will
have an addititnal 2 byte hole after the bitfields. This is
unsignificant however, and it doesn't influence gfar_priv's
size, because the whole structure is padded to be a 32B multiple.

Signed-off-by: Claudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agogianfar: Add device ref (dev) in gfar_private
Claudiu Manoil [Thu, 14 Feb 2013 05:00:02 +0000 (05:00 +0000)]
gianfar: Add device ref (dev) in gfar_private

Use device pointer (dev) to simplify the code and to
avoid double indirections, especially on the hot path.

Basically, instead of accessing priv to get the ofdev
reference and then accessing the ofdev structure to
dereference the needed dev pointer, we will get the
dev pointer directly from priv.

The dev pointer is required on the hot path, see gfar_new_rxbdp
or gfar_clean_rx_ring (or xmit), and this patch makes
it available directly from priv's 1st cacheline.

This change is reflected at asm level too, taking (the hot)
gfar_new_rxbdp():
initial version -
    18c0: 7c 7e 1b 78  mr      r30,r3

    18d0: 81 69 04 3c  lwz     r11,1084(r9)

    18d8: 34 6b 00 10  addic.  r3,r11,16
    18dc: 41 82 00 08  beq-    18e4

patched version -
    18d0: 80 69 04 38  lwz     r3,1080(r9)

    18d8: 2f 83 00 00  cmpwi   cr7,r3,0
    18dc: 41 9e 00 08  beq-    cr7,18e4

Signed-off-by: Claudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agogianfar: Remove unused device_node ref in gfar_private
Claudiu Manoil [Thu, 14 Feb 2013 05:00:01 +0000 (05:00 +0000)]
gianfar: Remove unused device_node ref in gfar_private

Remove unused device node pointer.
Remove duplicated SET_NETDEV_DEV().

Signed-off-by: Claudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec...
David S. Miller [Thu, 14 Feb 2013 18:29:20 +0000 (13:29 -0500)]
Merge branch 'master' of git://git./linux/kernel/git/klassert/ipsec-next

Steffen Klassert says:

====================
1) Remove a duplicated call to skb_orphan() in pf_key, from Cong Wang.

2) Prepare xfrm and pf_key for algorithms without pf_key support,
   from Jussi Kivilinna.

3) Fix an unbalanced lock in xfrm_output_one(), from Li RongQing.

4) Add an IPsec state resolution packet queue to handle
   packets that are send before the states are resolved.

5) xfrm4_policy_fini() is unused since 2.6.11, time to remove it.
   From Michal Kubecek.

6) The xfrm gc threshold was configurable just in the initial
   namespace, make it configurable in all namespaces. From
   Michal Kubecek.

7) We currently can not insert policies with mark and mask
   such that some flows would be matched from both policies.
   Allow this if the priorities of these policies are different,
   the one with the higher priority is used in this case.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: make ifla_br_policy and br_af_ops static
Cong Wang [Wed, 13 Feb 2013 19:57:12 +0000 (19:57 +0000)]
bridge: make ifla_br_policy and br_af_ops static

They are only used within this file.

Cc: Vlad Yasevich <vyasevic@redhat.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <amwang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobgmac: add read of interrupt mask after disabling interrupts
Nathan Hintz [Wed, 13 Feb 2013 19:14:10 +0000 (19:14 +0000)]
bgmac: add read of interrupt mask after disabling interrupts

The specs prescribe an immediate read of the interrupt mask after
disabling interrupts.  This patch updates the driver to match the
specs.

Signed-off-by: Nathan Hintz <nlhintz@hotmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: use __u16 in if_bridge.h
Cong Wang [Thu, 14 Feb 2013 05:32:31 +0000 (13:32 +0800)]
bridge: use __u16 in if_bridge.h

We should use "__u16" instead of "u16" in the user-space visable
header.

Cc: Vlad Yasevich <vyasevic@redhat.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <amwang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'bridge_vlan'
David S. Miller [Thu, 14 Feb 2013 00:43:18 +0000 (19:43 -0500)]
Merge branch 'bridge_vlan'

Vlad Yasevich says:

====================
VLAN filtering/VLAN aware bridge

Changes since v10
* Updated implemenation of ndo_fdb_del in emulex and qlogic drivers.

Changes since v9:
* series re-ordering so make functionality more distinct.  Basic vlan
  filtering is patches 1-4.  Support for PVID/untagged vlans is patches
  5 and 6.  VLAN support for FDB/MDB is patches 7-11.  Patch 12 is
  still additional egress policy.
* Slight simplification to code that extracts the VID from skb.  Since we
  now depend on the vlan module, at the time of input skb_tci is guaranteed
  to be set if the packet had 8021q header.  We can simply refere to it.
* Changed the opaque 'parent' pointer from prior patches to a union so we
  can be much more explicit in our assignments.
* Lots of additional testing with STP turned on.  No issues were observed.

Changes since v8:
* Unified vlans_to_* calls into a single interface
* Fixed the rest of the issues report by Michal Miroslaw
* Fixed a bug where fdb entries were not created for all added vlans.

Changes since v7:
* Rebases on the latest net-next and removed the vlan wrapper patch from
the series.
* Fixed a crash in br_fdb_add/br_fdb_delete.

Changes since v6:
* VLANs are now stored in a VLAN bitmap per port.  This allows for O(1)
lookup at ingress and egress.  We simply check to see if the bit associated
with the vlan id is set in the map.  The drawback to this approach is that
it wastes some space when there is only a small number of VLANs.
* In addition to the build time configuration option, VLAN filtering also has
a configuration paramter in sysfs.  By default the filtering is turned off
and all traffic is permitted.  When the filtring is turned on, we do strict
matching to the filter configured.  Thus, if there is no configuration, all
packets are rejected.  This was done to make the behavior more streight
forward.  Without this (and if egress policy patch is rejected), the
decision for how to forward untagged traffic that was not filtered at ingress
is almost impossible to make.  It would not be right to deliver to every
port that has PVID set as, each port may have a different PVID.
* Separate egress policy bitmap patch has been isolated and is provided last
in the series.  This has been a more contentious piece of functionality and I
wanted to isolate it so that it could easily be dropped and not block the whole
series.

Changes since v5:
 - Pulled VLAN filtering into its own file and made it a configuration options.
 - Made new vlan filtering option dependent on VLAN_8021Q.
 - Got rid of HW filter inlines and moved then vlan_core.c.
   (All of the above suggested by Stephen Hemminger)

Changes since v4:
 - Pull per-port vlan data into its own structures and give it to the bridge
   device thus making bridge device behave like a regular port for vlan
   configuration.
 - Add a per-vlan 'untagged' bitmap that determins egress policy.  If a port
   is part of this bitmap, traffic egresses untagged.
 - PVID is now used for ingress policy only.  Incomming frames without VLAN tag
   are assigned to the PVID vlan.  Egress is determined via bitmap memberships.
 - Allow for incremental config of a vlan.  Now, PVID and untagged memberships
   may be set on existing vlans.  They however can NOT be cleared separately.
 - VLAN deletion is now done via RTM_DELLINK command for PF_BRIDGE family.
   This cleans up the netlink interface.

Changes since v3:
 - Re-integrated compiler problems that got left out last time.  Appologies.
 - checkpatches.pl errors fixed

Changes since v2:
 - Added inline functiosn to manimulate vlan hw filters and re-use in 8021q
   and bridge code.
 - Use rtnl_dereference (Michael Tsirkin)
 - Remove synchronize_net() call (Eric Dumazet)
 - Fix NULL ptr deref bug I introduced in br_ifinfo_notify.

Changes since v1:
 - Fixed some forwarding bugs.
 - Add vlan to local fdb entries.  New local entries are created per vlan
   to facilite correct forwarding to bridge interface.
 - Allow configuration of vlans directly on the bridge master device
   in addition to ports.

Changes since rfc v2:
 - Per-port vlan bitmap is gone and is replaced with a vlan list.
 - Added bridge vlan list, which is referenced by each port.  Entries in
   the birdge vlan list have port bitmap that shows which port are parts
   of which vlan.
 - Netlink API changes.
 - Dropped sysfs support for now.  If people think this is really usefull,
   can add it back.
 - Support for native/untagged vlans.

Changes since rfc v1:
 - Comments addressed regarding formatting and RCU usage
 - iocts have been removed and changed over the netlink interface.
 - Added support of user added ndb entries.
 - changed sysfs interface to export a bitmap.  Also added a write interface.
   I am not sure how much I like it, but it made my testing easier/faster.  I
   might change the write interface to take text instead of binary.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: Separate egress policy bitmap
Vlad Yasevich [Wed, 13 Feb 2013 12:00:20 +0000 (12:00 +0000)]
bridge: Separate egress policy bitmap

Add an ability to configure a separate "untagged" egress
policy to the VLAN information of the bridge.  This superseeds PVID
policy and makes PVID ingress-only.  The policy is configured with a
new flag and is represented as a port bitmap per vlan.  Egress frames
with a VLAN id in "untagged" policy bitmap would egress
the port without VLAN header.

Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: Add vlan support for local fdb entries
Vlad Yasevich [Wed, 13 Feb 2013 12:00:19 +0000 (12:00 +0000)]
bridge: Add vlan support for local fdb entries

When VLAN is added to the port, a local fdb entry for that port
(the entry with the mac address of the port) is added for that
VLAN.  This way we can correctly determine if the traffic
is for the bridge itself.  If the address of the port changes,
we try to change all the local fdb entries we have for that port.

Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: Add vlan support to static neighbors
Vlad Yasevich [Wed, 13 Feb 2013 12:00:18 +0000 (12:00 +0000)]
bridge: Add vlan support to static neighbors

When a user adds bridge neighbors, allow him to specify VLAN id.
If the VLAN id is not specified, the neighbor will be added
for VLANs currently in the ports filter list.  If no VLANs are
configured on the port, we use vlan 0 and only add 1 entry.

Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Acked-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: Add vlan id to multicast groups
Vlad Yasevich [Wed, 13 Feb 2013 12:00:17 +0000 (12:00 +0000)]
bridge: Add vlan id to multicast groups

Add vlan_id to multicasts groups so that we know which vlan
each group belongs to and can correctly forward to appropriate vlan.

Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: Add vlan to unicast fdb entries
Vlad Yasevich [Wed, 13 Feb 2013 12:00:16 +0000 (12:00 +0000)]
bridge: Add vlan to unicast fdb entries

This patch adds vlan to unicast fdb entries that are created for
learned addresses (not the manually configured ones).  It adds
vlan id into the hash mix and uses vlan as an addditional parameter
for an entry match.

Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: Add the ability to configure pvid
Vlad Yasevich [Wed, 13 Feb 2013 12:00:15 +0000 (12:00 +0000)]
bridge: Add the ability to configure pvid

A user may designate a certain vlan as PVID.  This means that
any ingress frame that does not contain a vlan tag is assigned to
this vlan and any forwarding decisions are made with this vlan in mind.

Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: Implement vlan ingress/egress policy with PVID.
Vlad Yasevich [Wed, 13 Feb 2013 12:00:14 +0000 (12:00 +0000)]
bridge: Implement vlan ingress/egress policy with PVID.

At ingress, any untagged traffic is assigned to the PVID.
Any tagged traffic is filtered according to membership bitmap.

At egress, if the vlan matches the PVID, the frame is sent
untagged.  Otherwise the frame is sent tagged.

Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: Dump vlan information from a bridge port
Vlad Yasevich [Wed, 13 Feb 2013 12:00:13 +0000 (12:00 +0000)]
bridge: Dump vlan information from a bridge port

Using the RTM_GETLINK dump the vlan filter list of a given
bridge port.  The information depends on setting the filter
flag similar to how nic VF info is dumped.

Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: Add netlink interface to configure vlans on bridge ports
Vlad Yasevich [Wed, 13 Feb 2013 12:00:12 +0000 (12:00 +0000)]
bridge: Add netlink interface to configure vlans on bridge ports

Add a netlink interface to add and remove vlan configuration on bridge port.
The interface uses the RTM_SETLINK message and encodes the vlan
configuration inside the IFLA_AF_SPEC.  It is possble to include multiple
vlans to either add or remove in a single message.

Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: Verify that a vlan is allowed to egress on given port
Vlad Yasevich [Wed, 13 Feb 2013 12:00:11 +0000 (12:00 +0000)]
bridge: Verify that a vlan is allowed to egress on given port

When bridge forwards a frame, make sure that a frame is allowed
to egress on that port.

Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: Validate that vlan is permitted on ingress
Vlad Yasevich [Wed, 13 Feb 2013 12:00:10 +0000 (12:00 +0000)]
bridge: Validate that vlan is permitted on ingress

When a frame arrives on a port or transmitted by the bridge,
if we have VLANs configured, validate that a given VLAN is allowed
to enter the bridge.

Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: Add vlan filtering infrastructure
Vlad Yasevich [Wed, 13 Feb 2013 12:00:09 +0000 (12:00 +0000)]
bridge: Add vlan filtering infrastructure

Adds an optional infrustructure component to bridge that would allow
native vlan filtering in the bridge.  Each bridge port (as well
as the bridge device) now get a VLAN bitmap.  Each bit in the bitmap
is associated with a vlan id.  This way if the bit corresponding to
the vid is set in the bitmap that the packet with vid is allowed to
enter and exit the port.

Write access the bitmap is protected by RTNL and read access
protected by RCU.

Vlan functionality is disabled by default.

Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: sctp: add build check for sctp_sf_eat_sack_6_2/jsctp_sf_eat_sack
Daniel Borkmann [Wed, 13 Feb 2013 01:03:50 +0000 (01:03 +0000)]
net: sctp: add build check for sctp_sf_eat_sack_6_2/jsctp_sf_eat_sack

In order to avoid any future surprises of kernel panics due to jprobes
function mismatches (as e.g. fixed in 4cb9d6eaf85ecd: sctp: jsctp_sf_eat_sack:
fix jprobes function signature mismatch), we should check both function
types during build and scream loudly if they do not match. __same_type
resolves to __builtin_types_compatible_p, which is 1 in case both types
are the same and 0 otherwise, qualifiers are ignored. Tested by myself.

Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: sctp: minor: make jsctp_sf_eat_sack static
Daniel Borkmann [Wed, 13 Feb 2013 01:03:49 +0000 (01:03 +0000)]
net: sctp: minor: make jsctp_sf_eat_sack static

The function jsctp_sf_eat_sack can be made static, no need to extend
its visibility.

Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobgmac: return error on failed PHY write
Rafał Miłecki [Tue, 12 Feb 2013 23:14:51 +0000 (23:14 +0000)]
bgmac: return error on failed PHY write

Some callers may want to know if PHY write succeed. Also make PHY
functions static, they are not exported anywhere.

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobe2net: remove BUG_ON() in be_mcc_compl_is_new()
Sathya Perla [Tue, 12 Feb 2013 23:05:19 +0000 (23:05 +0000)]
be2net: remove BUG_ON() in be_mcc_compl_is_new()

The current code expects that the last word (with valid bit)
of an MCC compl is DMAed in one shot. This may not be the case.
Remove this assertion.

Signed-off-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: ethernet: ti: remove redundant NULL check.
Cyril Roelandt [Tue, 12 Feb 2013 12:52:30 +0000 (12:52 +0000)]
net: ethernet: ti: remove redundant NULL check.

cpdma_chan_destroy() on a NULL pointer is a no-op, so the NULL check in
cpdma_ctlr_destroy() can safely be removed.

Signed-off-by: Cyril Roelandt <tipecaml@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: Fix possible wrong checksum generation.
Pravin B Shelar [Mon, 11 Feb 2013 09:27:41 +0000 (09:27 +0000)]
net: Fix possible wrong checksum generation.

Patch cef401de7be8c4e (net: fix possible wrong checksum
generation) fixed wrong checksum calculation but it broke TSO by
defining new GSO type but not a netdev feature for that type.
net_gso_ok() would not allow hardware checksum/segmentation
offload of such packets without the feature.

Following patch fixes TSO and wrong checksum. This patch uses
same logic that Eric Dumazet used. Patch introduces new flag
SKBTX_SHARED_FRAG if at least one frag can be modified by
the user. but SKBTX_SHARED_FRAG flag is kept in skb shared
info tx_flags rather than gso_type.

tx_flags is better compared to gso_type since we can have skb with
shared frag without gso packet. It does not link SHARED_FRAG to
GSO, So there is no need to define netdev feature for this.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'tcp_tsoffset'
David S. Miller [Wed, 13 Feb 2013 18:22:24 +0000 (13:22 -0500)]
Merge branch 'tcp_tsoffset'

Andrey Vagin says:

====================
If a TCP socket will get live-migrated from one box to another the
timestamps (which are typically ON) will get screwed up -- the new
kernel will generate TS values that has nothing to do with what they
were on dump. The solution is to yet again fix the kernel and put a
"timestamp offset" on a socket.

A socket offset is added in places where externally visible tcp
timestamp option is parsed/initialized.

Connections in the SYN_RECV state are not supported, global
tcp_time_stamp is used for them, because repair mode doesn't support
this state. In a future it can be implemented by the similar way as for
TIME_WAIT sockets.

For time-wait sockets offset is inhereted by a proper tcp_sock.

A per-socket offset can be set only for sockets in repair mode.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agotcp: send packets with a socket timestamp
Andrey Vagin [Mon, 11 Feb 2013 05:50:19 +0000 (05:50 +0000)]
tcp: send packets with a socket timestamp

A socket timestamp is a sum of the global tcp_time_stamp and
a per-socket offset.

A socket offset is added in places where externally visible
tcp timestamp option is parsed/initialized.

Connections in the SYN_RECV state are not supported, global
tcp_time_stamp is used for them, because repair mode doesn't support
this state. In a future it can be implemented by the similar way
as for TIME_WAIT sockets.

Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Cc: James Morris <jmorris@namei.org>
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Cc: Patrick McHardy <kaber@trash.net>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agotcp: set and get per-socket timestamp
Andrey Vagin [Mon, 11 Feb 2013 05:50:18 +0000 (05:50 +0000)]
tcp: set and get per-socket timestamp

A timestamp can be set, only if a socket is in the repair mode.

This patch adds a new socket option TCP_TIMESTAMP, which allows to
get and set current tcp times stamp.

Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Cc: James Morris <jmorris@namei.org>
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Cc: Patrick McHardy <kaber@trash.net>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agotcp: adding a per-socket timestamp offset
Andrey Vagin [Mon, 11 Feb 2013 05:50:17 +0000 (05:50 +0000)]
tcp: adding a per-socket timestamp offset

This functionality is used for restoring tcp sockets. A tcp timestamp
depends on how long a system has been running, so it's differ for each
host. The solution is to set a per-socket offset.

A per-socket offset for a TIME_WAIT socket is inherited from a proper
tcp socket.

tcp_request_sock doesn't have a timestamp offset, because the repair
mode for them are not implemented.

Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Cc: James Morris <jmorris@namei.org>
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Cc: Patrick McHardy <kaber@trash.net>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'gfar-ethtool-atomic' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Wed, 13 Feb 2013 18:18:20 +0000 (13:18 -0500)]
Merge branch 'gfar-ethtool-atomic' of git://git./linux/kernel/git/paulg/linux

Paul Gortmaker says:

====================
Eric noticed that the handling of local u64 ethtool counters for
this driver commonly found on Freescale ppc-32 boards was racy.

However, before converting them over to atomic64_t, I noticed
that an internal struct was being used to determine the offsets
for exporting this data into the ethtool buffer, and in doing
so, it assumed that the counters would always be u64.  Rather
than keep this implicit assumption, a simple code cleanup gets
rid of the struct completely, and leaves less conversion sites.

The alternative solution would have been to take advantage of
the fact that the counters are all relating to error conditions,
and hence make them internally u32.  In doing so, we'd be assuming
that U32_MAX of any particular error condition is highly unlikely.
This might have made sense if any increments were in a hot path.

Tested with "ethtool -S eth0" on sbc8548 board.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonetpoll: fix smatch warnings in netpoll core code
Neil Horman [Wed, 13 Feb 2013 16:32:42 +0000 (11:32 -0500)]
netpoll: fix smatch warnings in netpoll core code

Dan Carpenter contacted me with some notes regarding some smatch warnings in the
netpoll code, some of which I introduced with my recent netpoll locking fixes,
some which were there prior.   Specifically they were:

net-next/net/core/netpoll.c:243 netpoll_poll_dev() warn: inconsistent
  returns mutex:&ni->dev_lock: locked (213,217) unlocked (210,243)
net-next/net/core/netpoll.c:706 netpoll_neigh_reply() warn: potential
  pointer math issue ('skb_transport_header(send_skb)' is a 128 bit pointer)

This patch corrects the locking imbalance (the first error), and adds some
parenthesis to correct the second error.  Tested by myself. Applies to net-next

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Dan Carpenter <dan.carpenter@oracle.com>
CC: "David S. Miller" <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: skbuff: fix compile error in skb_panic()
James Hogan [Wed, 13 Feb 2013 11:20:27 +0000 (11:20 +0000)]
net: skbuff: fix compile error in skb_panic()

I get the following build error on next-20130213 due to the following
commit:

commit f05de73bf82fbbc00265c06d12efb7273f7dc54a ("skbuff: create
skb_panic() function and its wrappers").

It adds an argument called panic to a function that uses the BUG() macro
which tries to call panic, but the argument masks the panic() function
declaration, resulting in the following error (gcc 4.2.4):

net/core/skbuff.c In function 'skb_panic':
net/core/skbuff.c +126 : error: called object 'panic' is not a function

This is fixed by renaming the argument to msg.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Jean Sacren <sakiwit@gmail.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agogianfar: convert u64 status counters to atomic64_t
Paul Gortmaker [Tue, 12 Feb 2013 20:38:19 +0000 (15:38 -0500)]
gianfar: convert u64 status counters to atomic64_t

While looking at some asm dump for an unrelated change, Eric
noticed in the following stats count increment code:

    50b8:       81 3c 01 f8     lwz     r9,504(r28)
    50bc:       81 5c 01 fc     lwz     r10,508(r28)
    50c0:       31 4a 00 01     addic   r10,r10,1
    50c4:       7d 29 01 94     addze   r9,r9
    50c8:       91 3c 01 f8     stw     r9,504(r28)
    50cc:       91 5c 01 fc     stw     r10,508(r28)

that a 64 bit counter was used on ppc-32 without sync
and hence the "ethtool -S" output was racy.

Here we convert all the values to use atomic64_t so that
the output will always be consistent.

Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
11 years agogianfar: remove largely unused gfar_stats struct
Paul Gortmaker [Tue, 12 Feb 2013 20:28:35 +0000 (15:28 -0500)]
gianfar: remove largely unused gfar_stats struct

The gfar_stats struct is only used in copying out data
via ethtool.  It is declared as the extra stats, followed
by the rmon stats.  However, the rmon stats are never
actually ever used in the driver; instead the rmon data
is a u32 register read that is cast directly into the
ethtool buf.

It seems the only reason rmon is in the struct at all is
to give the offset(s) at which it should be exported into
the ethtool buffer.  But note gfar_stats doesn't contain
a gfar_extra_stats as a substruct -- instead it contains
a u64 array of equal element count.  This implicitly means
we have two independent declarations of what gfar_extra_stats
really is.  Rather than have this duality, we already have
defines which give us the offset directly, and hence do not
need the struct at all.

Further, since we know the extra_stats is unconditionally
always present, we can write it out to the ethtool buf
1st, and then optionally write out the rmon data.  There
is no need for two independent loops, both of which are
simply copying out the extra_stats to buf offset zero.

This also helps pave the way towards allowing the extra
stats fields to be converted to atomic64_t values, without
having their types directly influencing the ethtool stats
export code (gfar_fill_stats) that expects to deal with u64.

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
11 years agoact_police: improved accuracy at high rates
Jiri Pirko [Tue, 12 Feb 2013 00:12:07 +0000 (00:12 +0000)]
act_police: improved accuracy at high rates

Current act_police uses rate table computed by the "tc" userspace
program, which has the following issue:

The rate table has 256 entries to map packet lengths to token (time
units).  With TSO sized packets, the 256 entry granularity leads to
loss/gain of rate, making the token bucket inaccurate.

Thus, instead of relying on rate table, this patch explicitly computes
the time and accounts for packet transmission times with nanosecond
granularity.

This is a followup to 56b765b79e9a78dc7d3f8850ba5e5567205a3ecd
("htb: improved accuracy at high rates").

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoact_police: move struct tcf_police to act_police.c
Jiri Pirko [Tue, 12 Feb 2013 00:12:06 +0000 (00:12 +0000)]
act_police: move struct tcf_police to act_police.c

It's not used anywhere else, so move it.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agotbf: improved accuracy at high rates
Jiri Pirko [Tue, 12 Feb 2013 00:12:05 +0000 (00:12 +0000)]
tbf: improved accuracy at high rates

Current TBF uses rate table computed by the "tc" userspace program,
which has the following issue:

The rate table has 256 entries to map packet lengths to
token (time units).  With TSO sized packets, the 256 entry granularity
leads to loss/gain of rate, making the token bucket inaccurate.

Thus, instead of relying on rate table, this patch explicitly computes
the time and accounts for packet transmission times with nanosecond
granularity.

This is a followup to 56b765b79e9a78dc7d3f8850ba5e5567205a3ecd
("htb: improved accuracy at high rates").

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agosch_api: introduce qdisc_watchdog_schedule_ns()
Jiri Pirko [Tue, 12 Feb 2013 00:12:04 +0000 (00:12 +0000)]
sch_api: introduce qdisc_watchdog_schedule_ns()

tbf will need to schedule watchdog in ns. No need to convert it twice.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agosch: make htb_rate_cfg and functions around that generic
Jiri Pirko [Tue, 12 Feb 2013 00:12:03 +0000 (00:12 +0000)]
sch: make htb_rate_cfg and functions around that generic

As it is going to be used in tbf as well, push these to generic code.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agohtb: initialize cl->tokens and cl->ctokens correctly
Jiri Pirko [Tue, 12 Feb 2013 00:12:02 +0000 (00:12 +0000)]
htb: initialize cl->tokens and cl->ctokens correctly

These are in ns so convert from ticks to ns.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agohtb: remove pointless first initialization of buffer and cbuffer
Jiri Pirko [Tue, 12 Feb 2013 00:12:01 +0000 (00:12 +0000)]
htb: remove pointless first initialization of buffer and cbuffer

These are initialized correctly a couple of lines later in the
function.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agohtb: use PSCHED_TICKS2NS()
Jiri Pirko [Tue, 12 Feb 2013 00:11:59 +0000 (00:11 +0000)]
htb: use PSCHED_TICKS2NS()

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Tue, 12 Feb 2013 23:58:28 +0000 (18:58 -0500)]
Merge git://git./linux/kernel/git/davem/net

Conflicts:
drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c

The bnx2x gso_type setting bug fix in 'net' conflicted with
changes in 'net-next' that broke the gso_* setting logic
out into a seperate function, which also fixes the bug in
question.  Thus, use the 'net-next' version.

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agohtb: fix values in opt dump
Jiri Pirko [Tue, 12 Feb 2013 00:12:00 +0000 (00:12 +0000)]
htb: fix values in opt dump

in htb_change_class() cl->buffer and cl->buffer are stored in ns.
So in dump, convert them back to psched ticks.

Note this was introduced by:
commit 56b765b79e9a78dc7d3f8850ba5e5567205a3ecd
    htb: improved accuracy at high rates

Please consider this for -net/-stable.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: fec_mpc52xx: Read MAC address from device-tree
Stefan Roese [Tue, 12 Feb 2013 03:05:38 +0000 (03:05 +0000)]
net: fec_mpc52xx: Read MAC address from device-tree

Until now, the MPC5200 FEC ethernet driver relied upon the bootloader
(U-Boot) to write the MAC address into the ethernet controller
registers. The Linux driver should not rely on such a thing. So
lets read the MAC address from the DT as it should be done here.

The following priority is now used to read the MAC address:

1) First, try OF node MAC address, if not present or invalid, then:

2) Read from MAC address registers, if invalid, then:

3) Log a warning message, and choose a random MAC address.

This fixes a problem with a MPC5200 board that uses the SPL U-Boot
version without FEC initialization before Linux booting for
boot speedup.

Additionally a status line is now be printed upon successful
driver probing, also displaying this MAC address.

Signed-off-by: Stefan Roese <sr@denx.de>
Cc: Anatolij Gustschin <agust@denx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agocxgb4vf: Fix VLAN extraction counter increment
Vipul Pandya [Tue, 12 Feb 2013 00:36:21 +0000 (00:36 +0000)]
cxgb4vf: Fix VLAN extraction counter increment

Signed-off-by: Santosh Rastapur <santosh@chelsio.com>
Signed-off-by: Vipul Pandya <vipul@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agodriver: net: ethernet: cpsw: dual emac interface implementation
Mugunthan V N [Mon, 11 Feb 2013 09:52:20 +0000 (09:52 +0000)]
driver: net: ethernet: cpsw: dual emac interface implementation

The CPSW switch can act as Dual EMAC by segregating the switch ports
using VLAN and port VLAN as per the TRM description in
14.3.2.10.2 Dual Mac Mode

Following CPSW components will be common for both the interfaces.
* Interrupt source is common for both eth interfaces
* Interrupt pacing is common for both interfaces
* Hardware statistics is common for all the ports
* CPDMA is common for both eth interface
* CPTS is common for both the interface and it should not be enabled on
  both the interface as timestamping information doesn't contain port
  information.

Constrains
* Reserved VID of One port should not be used in other interface which will
  enable switching functionality
* Same VID must not be used in both the interface which will enable switching
  functionality

Signed-off-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agodriver: net: ethernet: cpsw: make cpts as pointer
Mugunthan V N [Mon, 11 Feb 2013 09:52:19 +0000 (09:52 +0000)]
driver: net: ethernet: cpsw: make cpts as pointer

As CPTS is common module for both EMAC in Dual EMAC mode so making cpts as
pointer.

Signed-off-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agodriver: net: ethernet: davinci_cpdma: add support for directed packet and source...
Mugunthan V N [Mon, 11 Feb 2013 09:52:18 +0000 (09:52 +0000)]
driver: net: ethernet: davinci_cpdma: add support for directed packet and source port detection

* Introduced parameter to add port number for directed packet in cpdma_chan_submit
* Source port detection macro with DMA descriptor status

Signed-off-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'for-davem' of git://git.kernel.org/pub/scm/linux/kernel/git/linville...
David S. Miller [Tue, 12 Feb 2013 21:11:09 +0000 (16:11 -0500)]
Merge branch 'for-davem' of git://git./linux/kernel/git/linville/wireless

John W. Linville says:

====================
Here is another handful of late-breaking fixes intended for the 3.8
stream...  Hopefully the will still make it! :-)

There are three mac80211 fixes pulled from Johannes:

"Here are three fixes still for the 3.8 stream, the fix from Cong Ding
for the bad sizeof (Stephen Hemminger had pointed it out before but I'd
promptly forgotten), a mac80211 managed-mode channel context usage fix
where a downgrade would never stop until reaching non-HT and a bug in
the channel determination that could cause invalid channels like HT40+
on channel 11 to be used."

Also included is a mwl8k fix that avoids an oops when using mwl8k
devices that only support the 5 GHz band.

Please let me know if there are problems!
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoixgbe: Only set gso_type to SKB_GSO_TCPV4 as RSC does not support IPv6
Alexander Duyck [Tue, 12 Feb 2013 09:45:44 +0000 (09:45 +0000)]
ixgbe: Only set gso_type to SKB_GSO_TCPV4 as RSC does not support IPv6

The original fix that was applied for setting gso_type required more change
than necessary because it was assumed ixgbe does RSC on IPv6 frames and this
is not correct.  RSC is only supported with IPv4/TCP frames only.  As such we
can simplify the fix and avoid the unnecessary move of eth_type_trans.

The previous patch "ixgbe: fix gso type" and this patch reduce the entire fix
to one line that sets gso_type to TCPV4 if the frame is RSC.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: fix infinite loop in __skb_recv_datagram()
Eric Dumazet [Tue, 12 Feb 2013 06:16:53 +0000 (06:16 +0000)]
net: fix infinite loop in __skb_recv_datagram()

Tommi was fuzzing with trinity and reported the following problem :

commit 3f518bf745 (datagram: Add offset argument to __skb_recv_datagram)
missed that a raw socket receive queue can contain skbs with no payload.

We can loop in __skb_recv_datagram() with MSG_PEEK mode, because
wait_for_packet() is not prepared to skip these skbs.

[   83.541011] INFO: rcu_sched detected stalls on CPUs/tasks: {}
(detected by 0, t=26002 jiffies, g=27673, c=27672, q=75)
[   83.541011] INFO: Stall ended before state dump start
[  108.067010] BUG: soft lockup - CPU#0 stuck for 22s! [trinity-child31:2847]
...
[  108.067010] Call Trace:
[  108.067010]  [<ffffffff818cc103>] __skb_recv_datagram+0x1a3/0x3b0
[  108.067010]  [<ffffffff818cc33d>] skb_recv_datagram+0x2d/0x30
[  108.067010]  [<ffffffff819ed43d>] rawv6_recvmsg+0xad/0x240
[  108.067010]  [<ffffffff818c4b04>] sock_common_recvmsg+0x34/0x50
[  108.067010]  [<ffffffff818bc8ec>] sock_recvmsg+0xbc/0xf0
[  108.067010]  [<ffffffff818bf31e>] sys_recvfrom+0xde/0x150
[  108.067010]  [<ffffffff81ca4329>] system_call_fastpath+0x16/0x1b

Reported-by: Tommi Rantala <tt.rantala@gmail.com>
Tested-by: Tommi Rantala <tt.rantala@gmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: qmi_wwan: add Yota / Megafon M100-1 4g modem
Bjørn Mork [Tue, 12 Feb 2013 02:42:50 +0000 (02:42 +0000)]
net: qmi_wwan: add Yota / Megafon M100-1 4g modem

Interface layout:

 00 CD-ROM
 01 debug COM port
 02 AP control port
 03 modem
 04 usb-ethernet

Bus=01 Lev=02 Prnt=02 Port=01 Cnt=02 Dev#=  4 Spd=480  MxCh= 0
D:  Ver= 2.00 Cls=00(>ifc ) Sub=00 Prot=00 MxPS=64 #Cfgs=  1
P:  Vendor=0408 ProdID=ea42 Rev= 0.00
S:  Manufacturer=Qualcomm, Incorporated
S:  Product=Qualcomm CDMA Technologies MSM
S:  SerialNumber=353568051xxxxxx
C:* #Ifs= 5 Cfg#= 1 Atr=e0 MxPwr=500mA
I:* If#= 0 Alt= 0 #EPs= 2 Cls=08(stor.) Sub=06 Prot=50 Driver=usb-storage
E:  Ad=01(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E:  Ad=81(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
I:* If#= 1 Alt= 0 #EPs= 2 Cls=ff(vend.) Sub=ff Prot=ff Driver=(none)
E:  Ad=82(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E:  Ad=02(O) Atr=02(Bulk) MxPS= 512 Ivl=4ms
I:* If#= 2 Alt= 0 #EPs= 2 Cls=ff(vend.) Sub=ff Prot=ff Driver=(none)
E:  Ad=83(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E:  Ad=03(O) Atr=02(Bulk) MxPS= 512 Ivl=4ms
I:* If#= 3 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=ff Prot=ff Driver=(none)
E:  Ad=84(I) Atr=03(Int.) MxPS=  64 Ivl=2ms
E:  Ad=85(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E:  Ad=04(O) Atr=02(Bulk) MxPS= 512 Ivl=4ms
I:* If#= 4 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=ff Prot=ff Driver=(none)
E:  Ad=86(I) Atr=03(Int.) MxPS=  64 Ivl=2ms
E:  Ad=87(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E:  Ad=05(O) Atr=02(Bulk) MxPS= 512 Ivl=4ms

Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet: sctp: remove unused multiple cookie keys
Daniel Borkmann [Tue, 12 Feb 2013 05:15:33 +0000 (05:15 +0000)]
net: sctp: remove unused multiple cookie keys

Vlad says: The whole multiple cookie keys code is completely unused
and has been all this time. Noone uses anything other then the
secret_key[0] since there is no changeover support anywhere.

Thus, for now clean up its left-over fragments.

Cc: Neil Horman <nhorman@tuxdriver.com>
Cc: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Vlad Yasevich <vyasevich@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
John W. Linville [Tue, 12 Feb 2013 15:41:46 +0000 (10:41 -0500)]
Merge branch 'master' of git://git./linux/kernel/git/linville/wireless into for-davem

11 years agonetpoll: cleanup sparse warnings
Neil Horman [Mon, 11 Feb 2013 10:25:31 +0000 (10:25 +0000)]
netpoll: cleanup sparse warnings

With my recent commit I introduced two sparse warnings.  Looking closer there
were a few more in the same file, so I fixed them all up.  Basic rcu pointer
dereferencing suff.

I've validated these changes using CONFIG_PROVE_RCU while starting and stopping
netconsole repeatedly in bonded and non-bonded configurations

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: fengguang.wu@intel.com
CC: David Miller <davem@davemloft.net>
CC: eric.dumazet@gmail.com
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonetpoll: Fix __netpoll_rcu_free so that it can hold the rtnl lock
Neil Horman [Mon, 11 Feb 2013 10:25:30 +0000 (10:25 +0000)]
netpoll: Fix __netpoll_rcu_free so that it can hold the rtnl lock

__netpoll_rcu_free is used to free netpoll structures when the rtnl_lock is
already held.  The mechanism is used to asynchronously call __netpoll_cleanup
outside of the holding of the rtnl_lock, so as to avoid deadlock.
Unfortunately, __netpoll_cleanup modifies pointers (dev->np), which means the
rtnl_lock must be held while calling it.  Further, it cannot be held, because
rcu callbacks may be issued in softirq contexts, which cannot sleep.

Fix this by converting the rcu callback to a work queue that is guaranteed to
get scheduled in process context, so that we can hold the rtnl properly while
calling __netpoll_cleanup

Tested successfully by myself.

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: "David S. Miller" <davem@davemloft.net>
CC: Cong Wang <amwang@redhat.com>
CC: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoskbuff: create skb_panic() function and its wrappers
Jean Sacren [Mon, 11 Feb 2013 13:30:38 +0000 (13:30 +0000)]
skbuff: create skb_panic() function and its wrappers

Create skb_panic() function in lieu of both skb_over_panic() and
skb_under_panic() so that code duplication would be avoided. Update type
and variable name where necessary.

Jiri Pirko suggested using wrappers so that we would be able to keep the
fruits of the original code.

Signed-off-by: Jean Sacren <sakiwit@gmail.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agomwl8k: fix band for supported channels
Jonas Gorski [Fri, 8 Feb 2013 15:07:25 +0000 (16:07 +0100)]
mwl8k: fix band for supported channels

The band field for the supported channels were left unpopulated, making
them default to 0 == IEEE80211_BAND_2GHZ, even for the 5GHz channels.

This resulted in null pointer accesses if anything tries to access
wiphy->bands[channel->band] of a 5GHz channel on 5GHz only cards, since
wiphy->bands[2GHZ] is NULL for them (e.g. cfg80211_chandef_usable does).

Example kernel OOPS:

[  665.669993] Unable to handle kernel NULL pointer dereference at virtual address 00000016
[  665.678194] pgd = c6d58000
[  665.680941] [00000016] *pgd=06f8a831, *pte=00000000, *ppte=00000000
[  665.687303] Internal error: Oops: 17 [#1]
(...)
[  666.116373] Backtrace:
[  666.118866] [<bf0368dc>] (cfg80211_chandef_usable+0x0/0x1bc [cfg80211]) from [<bf025e64>] (nl80211_leave_mesh+0x244/0x264 [cfg80211])
[  666.130919]  r7:c6d12100 r6:0000143c r5:c0611c48 r4:c0611b98
[  666.136668] [<bf025d84>] (nl80211_leave_mesh+0x164/0x264 [cfg80211]) from [<bf02634c>] (nl80211_remain_on_channel+0x2a0/0x358 [cfg80211])
[  666.149074]  r7:c6d12000 r6:c6d12000 r5:c6f4f368 r4:00000003
[  666.154814] [<bf0262ec>] (nl80211_remain_on_channel+0x240/0x358 [cfg80211]) from [<bf02ddb0>] (nl80211_set_wiphy+0x264/0x560 [cfg80211])
[  666.167150] [<bf02db4c>] (nl80211_set_wiphy+0x0/0x560 [cfg80211]) from [<c01f94e0>] (genl_rcv_msg+0x1b8/0x1f8)
[  666.177205] [<c01f9328>] (genl_rcv_msg+0x0/0x1f8) from [<c01f89a0>] (netlink_rcv_skb+0x58/0xb4)
[  666.185949] [<c01f8948>] (netlink_rcv_skb+0x0/0xb4) from [<c01f931c>] (genl_rcv+0x20/0x2c)
[  666.194251]  r6:c6f70780 r5:0000002c r4:c6f70780 r3:00000001
[  666.199973] [<c01f92fc>] (genl_rcv+0x0/0x2c) from [<c01f8418>] (netlink_unicast+0x154/0x1f4)
[  666.208449]  r4:c785ea00 r3:c01f92fc
[  666.212057] [<c01f82c4>] (netlink_unicast+0x0/0x1f4) from [<c01f8790>] (netlink_sendmsg+0x230/0x2b0)
[  666.221240] [<c01f8560>] (netlink_sendmsg+0x0/0x2b0) from [<c01cccf8>] (sock_sendmsg+0x90/0xa4)
[  666.229986] [<c01ccc68>] (sock_sendmsg+0x0/0xa4) from [<c01cdcb0>] (__sys_sendmsg+0x290/0x298)
[  666.238637]  r9:00000000 r8:c0611ec8 r6:0000002c r5:c0610000 r4:c0611f64
[  666.245411] [<c01cda20>] (__sys_sendmsg+0x0/0x298) from [<c01cf52c>] (sys_sendmsg+0x44/0x6c)
[  666.253897] [<c01cf4e8>] (sys_sendmsg+0x0/0x6c) from [<c00090a0>] (ret_fast_syscall+0x0/0x2c)
[  666.262460]  r6:00000000 r5:beeff96c r4:00000005

Signed-off-by: Jonas Gorski <jogo@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
11 years agoMerge branch 'for-john' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211
John W. Linville [Mon, 11 Feb 2013 19:25:28 +0000 (14:25 -0500)]
Merge branch 'for-john' of git://git./linux/kernel/git/jberg/mac80211

11 years agobridge: set priority of STP packets
Stephen Hemminger [Mon, 11 Feb 2013 08:22:22 +0000 (08:22 +0000)]
bridge: set priority of STP packets

Spanning Tree Protocol packets should have always been marked as
control packets, this causes them to get queued in the high prirority
FIFO. As Radia Perlman mentioned in her LCA talk, STP dies if bridge
gets overloaded and can't communicate. This is a long-standing bug back
to the first versions of Linux bridge.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agomrp: make mrp_rcv static
Stephen Hemminger [Mon, 11 Feb 2013 06:25:24 +0000 (06:25 +0000)]
mrp: make mrp_rcv static

Sparse spotted local function that could be static.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoethtool: fix sparse warning
Stephen Hemminger [Mon, 11 Feb 2013 06:22:28 +0000 (06:22 +0000)]
ethtool: fix sparse warning

Fixes sparse complaints about dropping __user in casts.
 warning: cast removes address space of expression

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobridge: use dev->addr_assign_type to see if user change mac
Jiri Pirko [Sun, 10 Feb 2013 11:41:01 +0000 (11:41 +0000)]
bridge: use dev->addr_assign_type to see if user change mac

And remove no longer used br->flags.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoipv6: by default join ff01::1 and in case of forwarding ff01::2 and ff05:2
Hannes Frederic Sowa [Sun, 10 Feb 2013 03:50:18 +0000 (03:50 +0000)]
ipv6: by default join ff01::1 and in case of forwarding ff01::2 and ff05:2

Cc: Erik Hugne <erik.hugne@ericsson.com>
Cc: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoipv6: don't accept multicast traffic with scope 0
Hannes Frederic Sowa [Sun, 10 Feb 2013 05:35:22 +0000 (05:35 +0000)]
ipv6: don't accept multicast traffic with scope 0

v2:
a) moved before multicast source address check
b) changed comment to netdev style

Cc: Erik Hugne <erik.hugne@ericsson.com>
Cc: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Acked-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoipv6: don't let node/interface scoped multicast traffic escape on the wire
Hannes Frederic Sowa [Sun, 10 Feb 2013 02:33:35 +0000 (02:33 +0000)]
ipv6: don't let node/interface scoped multicast traffic escape on the wire

Reported-by: Erik Hugne <erik.hugne@ericsson.com>
Cc: Erik Hugne <erik.hugne@ericsson.com>
Cc: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoxfrm: Allow inserting policies with matching mark and different priorities
Steffen Klassert [Mon, 11 Feb 2013 06:02:36 +0000 (07:02 +0100)]
xfrm: Allow inserting policies with matching mark and different priorities

We currently can not insert policies with mark and mask
such that some flows would be matched from both policies.
We make this possible when the priority of these policies
are different. If both policies match a flow, the one with
the higher priority is used.

Reported-by: Emmanuel Thierry <emmanuel.thierry@telecom-bretagne.eu>
Reported-by: Romain Kuntz <r.kuntz@ipflavors.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
11 years agomac80211: fix channel selection bug
Johannes Berg [Sat, 9 Feb 2013 20:46:34 +0000 (21:46 +0100)]
mac80211: fix channel selection bug

When trying to connect to an AP that advertises HT but not
VHT, the mac80211 code erroneously uses the configuration
from the AP as is instead of checking it against regulatory
and local capabilities. This can lead to using an invalid
or even inexistent channel (like 11/HT40+).

Additionally, the return flags from downgrading must be
ORed together, to collect them from all of the downgrades.
Also clarify the message.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
11 years agoqlcnic: fix mailbox response handling
Jitendra Kalsaria [Sat, 9 Feb 2013 09:29:55 +0000 (09:29 +0000)]
qlcnic: fix mailbox response handling

o Do not read mailbox registers on timeout
o Add a helper function to handle mailbox response

Signed-off-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: enhance MSIX allocation failure log message
Manish Chopra [Sat, 9 Feb 2013 09:29:54 +0000 (09:29 +0000)]
qlcnic: enhance MSIX allocation failure log message

Signed-off-by: Manish Chopra <manish.chopra@qlogic.com>
Signed-off-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: fix async event handling during diagnostic loopback test
Jitendra Kalsaria [Sat, 9 Feb 2013 09:29:53 +0000 (09:29 +0000)]
qlcnic: fix async event handling during diagnostic loopback test

o Handle async events during diagnostic loopback test
o Clear loopback mode on failure to receive async events

Signed-off-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: refactor 83xx diagnostic IRQ test
Jitendra Kalsaria [Sat, 9 Feb 2013 09:29:52 +0000 (09:29 +0000)]
qlcnic: refactor 83xx diagnostic IRQ test

Cleanly separate 83xx diagnostic IRQ test from 82xx

Signed-off-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: refactor 83xx diagnostic loopback test
Jitendra Kalsaria [Sat, 9 Feb 2013 09:29:51 +0000 (09:29 +0000)]
qlcnic: refactor 83xx diagnostic loopback test

Cleanly separate 83xx diagnostic loopback test routines from 82xx

Signed-off-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: helper routine to handle async events
Jitendra Kalsaria [Sat, 9 Feb 2013 09:29:50 +0000 (09:29 +0000)]
qlcnic: helper routine to handle async events

Create a helper routine to handle async events, as it is being called
from multiple places

Signed-off-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: fix firmware based IDC participation
Jitendra Kalsaria [Sat, 9 Feb 2013 09:29:49 +0000 (09:29 +0000)]
qlcnic: fix firmware based IDC participation

Driver needs to stop participating in firmware based Inter Driver
Communication (IDC) while unloading driver

Signed-off-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: fix initialize NIC mailbox command
Himanshu Madhani [Sat, 9 Feb 2013 09:29:48 +0000 (09:29 +0000)]
qlcnic: fix initialize NIC mailbox command

Register for firmware based Inter Driver Communication (IDC) using initialize
NIC as the first mailbox command

Signed-off-by: Himanshu Madhani <himanshu.madhani@qlogic.com>
Signed-off-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet neighbour, decnet: Ensure to align device private data on preferred alignment.
YOSHIFUJI Hideaki / 吉藤英明 [Sat, 9 Feb 2013 07:00:59 +0000 (07:00 +0000)]
net neighbour, decnet: Ensure to align device private data on preferred alignment.

To allow both of protocol-specific data and device-specific data
attached with neighbour entry, and to eliminate size calculation
cost when allocating entry, sizeof protocol-speicic data must be
multiple of NEIGH_PRIV_ALIGN.  On 64bit archs,
sizeof(struct dn_neigh) is multiple of NEIGH_PRIV_ALIGN, but on
32bit archs, it was not.

Introduce NEIGH_ENTRY_SPACE() macro to ensure that protocol-specific
entry-size meets our requirement.

Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoipv6 mcast: Do not join device multicast for interface-local multicasts.
YOSHIFUJI Hideaki / 吉藤英明 [Sat, 9 Feb 2013 04:29:58 +0000 (04:29 +0000)]
ipv6 mcast: Do not join device multicast for interface-local multicasts.

RFC4291 (IPv6 addressing architecture) says that interface-Local scope
spans only a single interface on a node.  We should not join L2 device
multicast list for addresses in interface-local (or smaller) scope.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'master' of git://1984.lsi.us.es/nf
David S. Miller [Mon, 11 Feb 2013 01:44:08 +0000 (20:44 -0500)]
Merge branch 'master' of git://1984.lsi.us.es/nf

Pablo Neira Ayuso says:

====================
The following patchset contains Netfilter/IPVS fixes for 3.8-rc7, they are:

* Fix oops in IPVS state-sync due to releasing a random memory area due
  to unitialized pointer, from Dan Carpenter.

* Fix SCTP flow establishment due to bad checksumming mangling in IPVS,
  from Daniel Borkmann.

* Three fixes for the recently added IPv6 NPT, all from YOSHIFUJI Hideaki,
  with an amendment collapsed into those patches from Ulrich Weber. They
  fiix adjustment calculation, fix prefix mangling and ensure LSB of
  prefixes are zeroes (as required by RFC).

Specifically, it took me a while to validate the 1's complement arithmetics/
checksumming approach in the IPv6 NPT code.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoveth: fix NULL dereference in veth_dellink()
Eric Dumazet [Fri, 8 Feb 2013 20:10:49 +0000 (20:10 +0000)]
veth: fix NULL dereference in veth_dellink()

commit d0e2c55e7c940 (veth: avoid a NULL deref in veth_stats_one)
added another NULL deref in veth_dellink().

# ip link add name veth1 type veth peer name veth0
# rmmod veth

We crash because veth_dellink() is called twice, so we must
take care of NULL peer.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoarp: fix possible crash in arp_rcv()
Eric Dumazet [Fri, 8 Feb 2013 18:48:21 +0000 (18:48 +0000)]
arp: fix possible crash in arp_rcv()

We should call skb_share_check() before pskb_may_pull(), or we
can crash in pskb_expand_head()

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'mvrp'
David S. Miller [Mon, 11 Feb 2013 01:37:27 +0000 (20:37 -0500)]
Merge branch 'mvrp'

David Ward says:

====================
The Linux kernel currently implements the GARP VLAN Registration
Protocol (GVRP) from IEEE 802.1Q-1998 (applicant-only participant).
When the GVRP flag is set for a VLAN interface on a Linux host, the
host advertises its membership in the VLAN to the attached bridge/
switch, so that it is not necessary to manually configure the bridge/
switch port to participate in the VLAN.

GVRP has been superseded by the Multiple VLAN Registration Protocol
(MVRP) in IEEE 802.1Q-2011, which addresses scalability concerns about
the earlier protocol.  The following patches add support for MVRP to
the Linux kernel and iproute2 utility. They are based largely off of
the existing implementation of GVRP, but have been modified for the
new PDU structure and state machine.

This implementation was tested with two Juniper EX4200 switches.
====================

Signed-off-by: David Ward <david.ward@ll.mit.edu>
Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet/8021q: Implement Multiple VLAN Registration Protocol (MVRP)
David Ward [Fri, 8 Feb 2013 17:17:07 +0000 (17:17 +0000)]
net/8021q: Implement Multiple VLAN Registration Protocol (MVRP)

Initial implementation of the Multiple VLAN Registration Protocol
(MVRP) from IEEE 802.1Q-2011, based on the existing implementation
of the GARP VLAN Registration Protocol (GVRP).

Signed-off-by: David Ward <david.ward@ll.mit.edu>
Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet/802: Implement Multiple Registration Protocol (MRP)
David Ward [Fri, 8 Feb 2013 17:17:06 +0000 (17:17 +0000)]
net/802: Implement Multiple Registration Protocol (MRP)

Initial implementation of the Multiple Registration Protocol (MRP)
from IEEE 802.1Q-2011, based on the existing implementation of the
Generic Attribute Registration Protocol (GARP).

Signed-off-by: David Ward <david.ward@ll.mit.edu>
Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'gso_type'
David S. Miller [Mon, 11 Feb 2013 01:14:46 +0000 (20:14 -0500)]
Merge branch 'gso_type'

Michael S. Tsirkin says:

====================
At the moment, macvtap crashes are observed if macvtap is attached
to an interface with LRO enabled.
The crash in question is BUG() in macvtap_skb_to_vnet_hdr.
This happens because several drivers set gso_size but not gso_type
in incoming skbs.
This didn't use to be the case: with intel cards on 3.2 and older
kernels, with qlogic - on 3.4 and older kernels, so it's a regression if
not a recent one.
The following patches fix this for qlogic, broadcom and intel drivers.

I tested that the patch fixes the crash for ixgbe but
don't have qlogic/broadcom hardware to test.
I also only tested TCPv4.

Please review, and consider for 3.8.

Changes from v1:
- added missing htons as suggested by Eric
- backported the relevant bits from
  cbf1de72324a8105ddcc3d9ce9acbc613faea17e for bnx2x
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobnx2x: set gso_type
Michael S. Tsirkin [Thu, 7 Feb 2013 03:13:22 +0000 (03:13 +0000)]
bnx2x: set gso_type

In LRO mode, bnx2x set gso_size but not gso type.
This leads to crashes in macvtap.
Commit cbf1de72324a8105ddcc3d9ce9acbc613faea17e
queued for 3.9 includes a more complete fix.
This is a minimal patch to avoid the crash, for 3.8.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: set gso_type
Michael S. Tsirkin [Thu, 7 Feb 2013 03:13:13 +0000 (03:13 +0000)]
qlcnic: set gso_type

qlcnic set gso_size but not gso type. This leads to crashes
in macvtap.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoixgbe: fix gso type
Michael S. Tsirkin [Thu, 7 Feb 2013 03:13:05 +0000 (03:13 +0000)]
ixgbe: fix gso type

ixgbe set gso_size but not gso_type. This leads to
crashes in macvtap.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agostmmac: mdio register has to fail if the phy is not found
Giuseppe CAVALLARO [Wed, 6 Feb 2013 20:47:52 +0000 (20:47 +0000)]
stmmac: mdio register has to fail if the phy is not found

With this patch the stmmac fails in case of the phy device
is not found; w/o this fix the mdio can be register twice when
do down/up the iface and this is not correct.

Reported-by: Stas <stsp@list.ru>
Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agostmmac: fix macro used for debugging the xmit
Giuseppe CAVALLARO [Wed, 6 Feb 2013 20:47:51 +0000 (20:47 +0000)]
stmmac: fix macro used for debugging the xmit

This patch fixes the name of the macro used for
debugging the transmit process. I used STMMAC_TX_DEBUG
instead of STMMAC_XMIT_DEBUG.

Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'vsock'
David S. Miller [Mon, 11 Feb 2013 01:10:10 +0000 (20:10 -0500)]
Merge branch 'vsock'

Andy King says:

====================
In an effort to improve the out-of-the-box experience with Linux kernels for
VMware users, VMware is working on readying the VM Sockets (VSOCK, formerly
VMCI Sockets) (vsock) kernel module for inclusion in the Linux kernel. The
purpose of this post is to acquire feedback on the vsock kernel module.

Unlike previous submissions, where the new socket family was entirely reliant
on VMware's VMCI PCI device (and thus VMware's hypervisor), VM Sockets is now
completely[1] separated out into two parts, each in its own module:

o Core socket code, which is transport-neutral and invokes transport
  callbacks to communicate with the hypervisor.  This is vsock.ko.
o A VMCI transport, which communicates over VMCI with the VMware hypervisor.
  This is vmw_vsock_vmci_transport.ko, and it registers with the core module
  as a transport.

This should provide a path to introducing additional transports, for example
virtio, with the ultimate goal being to make this new socket family
hypervisor-neutral.

[1] If Gerd tries it and determines this to be false (still), I'll ship him
    a keg of beer.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoVSOCK: Introduce VM Sockets
Andy King [Wed, 6 Feb 2013 14:23:56 +0000 (14:23 +0000)]
VSOCK: Introduce VM Sockets

VM Sockets allows communication between virtual machines and the hypervisor.
User level applications both in a virtual machine and on the host can use the
VM Sockets API, which facilitates fast and efficient communication between
guest virtual machines and their host.  A socket address family, designed to be
compatible with UDP and TCP at the interface level, is provided.

Today, VM Sockets is used by various VMware Tools components inside the guest
for zero-config, network-less access to VMware host services.  In addition to
this, VMware's users are using VM Sockets for various applications, where
network access of the virtual machine is restricted or non-existent.  Examples
of this are VMs communicating with device proxies for proprietary hardware
running as host applications and automated testing of applications running
within virtual machines.

The VMware VM Sockets are similar to other socket types, like Berkeley UNIX
socket interface.  The VM Sockets module supports both connection-oriented
stream sockets like TCP, and connectionless datagram sockets like UDP. The VM
Sockets protocol family is defined as "AF_VSOCK" and the socket operations
split for SOCK_DGRAM and SOCK_STREAM.

For additional information about the use of VM Sockets, please refer to the
VM Sockets Programming Guide available at:

https://www.vmware.com/support/developer/vmci-sdk/

Signed-off-by: George Zhang <georgezhang@vmware.com>
Signed-off-by: Dmitry Torokhov <dtor@vmware.com>
Signed-off-by: Andy king <acking@vmware.com>
Signed-off-by: David S. Miller <davem@davemloft.net>