When it's in the waiting-for-ACK state, the AFS filesystem needs to check
the result of rxrpc_kernel_recv_data() any time it is notified to see if it
is indicating a fatal error.  If this is the case, it needs to mark the
call completed otherwise the call just sits there and never goes away.
Signed-off-by: David Howells <dhowells@redhat.com>

In rxrpc_kernel_recv_data(), when we return the error number incurred by a
failed call, we must negate it before returning it as it's stored as
positive (that's what we have to pass back to userspace).
Signed-off-by: David Howells <dhowells@redhat.com>

The call's background processor work item needs to notify the socket when
it completes a call so that recvmsg() or the AFS fs can deal with it.
Without this, call expiry isn't handled.
Signed-off-by: David Howells <dhowells@redhat.com>

When a call expires, it must be queued for the background processor to deal
with otherwise a service call that is improperly terminated will just sit
there awaiting an ACK and won't expire.
Signed-off-by: David Howells <dhowells@redhat.com>

OpenAFS doesn't always correctly terminate client calls that it makes -
this includes calls the OpenAFS servers make to the cache manager service.
It should end the client call with either:

 (1) An ACK that has firstPacket set to one greater than the seq number of
     the reply DATA packet with the LAST_PACKET flag set (thereby
     hard-ACK'ing all packets).  nAcks should be 0 and acks[] should be
     empty (ie. no soft-ACKs).

 (2) An ACKALL packet.

OpenAFS, though, may send an ACK packet with firstPacket set to the last
seq number or less and soft-ACKs listed for all packets up to and including
the last DATA packet.

The transmitter, however, is obliged to keep the call live and the
soft-ACK'd DATA packets around until they're hard-ACK'd as the receiver is
permitted to drop any merely soft-ACK'd packet and request retransmission
by sending an ACK packet with a NACK in it.

Further, OpenAFS will also terminate a client call by beginning the next
client call on the same connection channel.  This implicitly completes the
previous call.

This patch handles implicit ACK of a call on a channel by the reception of
the first packet of the next call on that channel.

If another call doesn't come along to implicitly ACK a call, then we have
to time the call out.  There are some bugs there that will be addressed in
subsequent patches.
Signed-off-by: David Howells <dhowells@redhat.com>

Separate the output of PING ACKs from the output of other sorts of ACK so
that if we receive a PING ACK and schedule transmission of a PING RESPONSE
ACK, the response doesn't get cancelled by a PING ACK we happen to be
scheduling transmission of at the same time.

If a PING RESPONSE gets lost, the other side might just sit there waiting
for it and refuse to proceed otherwise.
Signed-off-by: David Howells <dhowells@redhat.com>

Split rxrpc_send_data_packet() to separate ACK generation (which is more
complicated) from ABORT generation.  This simplifies the code a bit and
fixes the following warning:

In file included from ../net/rxrpc/output.c:20:0:
net/rxrpc/output.c: In function 'rxrpc_send_call_packet':
net/rxrpc/ar-internal.h:1187:27: error: 'top' may be used uninitialized in this function [-Werror=maybe-uninitialized]
net/rxrpc/output.c:103:24: note: 'top' was declared here
net/rxrpc/output.c:225:25: error: 'hard_ack' may be used uninitialized in this function [-Werror=maybe-uninitialized]
Reported-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David Howells <dhowells@redhat.com>

When a reply is deemed lost, we send a ping to find out the other end
received all the request data packets we sent.  This should be limited to
client calls and we shouldn't do this on service calls.
Signed-off-by: David Howells <dhowells@redhat.com>

If an call comes in to a local endpoint that isn't listening for any
incoming calls at the moment, an oops will happen.  We need to check that
the local endpoint's service pointer isn't NULL before we dereference it.
Signed-off-by: David Howells <dhowells@redhat.com>

Remove a duplicate const keyword.
Signed-off-by: David Howells <dhowells@redhat.com>

struct rxrpc_local->service is marked __rcu - this means that accesses of
it need to be managed using RCU wrappers.  There are two such places in
rxrpc_release_sock() where the value is checked and cleared.  Fix this by
using the appropriate wrappers.
Signed-off-by: David Howells <dhowells@redhat.com>

Gavin Shan says:

====================
net/ncsi: NCSI Improvment and bug fixes

This series of patches improves NCSI stack according to the comments
I received after the NCSI code was merged to 4.8.rc1:

  * PATCH[1/8] fixes the build warning caused by xchg() with ia64-linux-gcc.
    The atomic operations are removed. The NCSI's lock should be taken when
    reading or updating its state and chained state.
  * Channel ID (0x1f) is the reserved one and it cannot be valid channel ID.
    So we needn't try to probe channel whose ID is 0x1f. PATCH[2/8] and
    PATCH[3/8] are addressing this issue.
  * The request IDs are assigned in round-robin fashion, but it's broken.
    PATCH[4/8] make it work.
  * PATCH[5/8] and PATCH[6/8] reworks the channel monitoring to improve the
    code readability and its robustness.
  * PATCH[7/8] and PATCH[8/8] introduces ncsi_stop_dev() so that the network
    device can be closed and opened afterwards. No error will be seen.

Changelog
=========
v2:
  * The NCSI's lock is taken when reading or updating its state as the
    {READ,WRITE}_ONCE() isn't reliable.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

This stops NCSI device when closing the network device so that the
NCSI device can be reenabled later.
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

This introduces ncsi_stop_dev(), as counterpart to ncsi_start_dev(),
to stop the NCSI device so that it can be reenabled in future. This
API should be called when the network device driver is going to
shutdown the device. There are 3 things done in the function: Stop
the channel monitoring; Reset channels to inactive state; Report
NCSI link down.
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

The original NCSI channel monitoring was implemented based on a
backoff algorithm: the GLS response should be received in the
specified interval. Otherwise, the channel is regarded as dead
and failover should be taken if current channel is an active one.
There are several problems in the implementation: (A) On BCM5718,
we found when the IID (Instance ID) in the GLS command packet
changes from 255 to 1, the response corresponding to IID#1 never
comes in. It means we cannot make the unfair judgement that the
channel is dead when one response is missed. (B) The code's
readability should be improved. (C) We should do failover when
current channel is active one and the channel monitoring should
be marked as disabled before doing failover.

This reworks the channel monitoring to address all above issues.
The fields for channel monitoring is put into separate struct
and the state of channel monitoring is predefined. The channel
is regarded alive if the network controller responses to one of
two GLS commands or both of them in 5 seconds.
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

There is only one NCSI request property for now: the response for
the sent command need drive the workqueue or not. So we had one
field (@driven) for the purpose. We lost the flexibility to extend
NCSI request properties.

This replaces @driven with @flags and @req_flags in NCSI request
and NCSI command argument struct. Each bit of the newly introduced
field can be used for one property. No functional changes introduced.
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

The NCSI request index (struct ncsi_request::id) is put into instance
ID (IID) field while sending NCSI command packet. It was designed the
available IDs are given in round-robin fashion. @ndp->request_id was
introduced to represent the next available ID, but it has been used
as number of successively allocated IDs. It breaks the round-robin
design. Besides, we shouldn't put 0 to NCSI command packet's IID
field, meaning ID#0 should be reserved according section 6.3.1.1
in NCSI spec (v1.1.0).

This fixes above two issues. With it applied, the available IDs will
be assigned in round-robin fashion and ID#0 won't be assigned.
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

We needn't send CIS (Clear Initial State) command to the NCSI
reserved channel (0x1f) in the enumeration. We shouldn't receive
a valid response from CIS on NCSI channel 0x1f.
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

This defines NCSI_RESERVED_CHANNEL as the reserved NCSI channel
ID (0x1f). No logical changes introduced.
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

xchg() is used to set NCSI channel's state in order for consistent
access to the state. xchg()'s return value should be used. Otherwise,
one build warning will be raised (with -Wunused-value) as below message
indicates. It is reported by ia64-linux-gcc (GCC) 4.9.0.

 net/ncsi/ncsi-manage.c: In function 'ncsi_channel_monitor':
 arch/ia64/include/uapi/asm/cmpxchg.h:56:2: warning: value computed is \
 not used [-Wunused-value]
  ((__typeof__(*(ptr))) __xchg((unsigned long) (x), (ptr), sizeof(*(ptr))))
   ^
 net/ncsi/ncsi-manage.c:202:3: note: in expansion of macro 'xchg'
  xchg(&nc->state, NCSI_CHANNEL_INACTIVE);

This removes the atomic access to NCSI channel's state avoid the above
build warning. We have to hold the channel's lock when its state is readed
or updated. No functional changes introduced.
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

This is a respin of a patch to fix a relatively easily reproducible kernel
panic related to the all_adj_list handling for netdevs in recent kernels.

The following sequence of commands will reproduce the issue:

ip link add link eth0 name eth0.100 type vlan id 100
ip link add link eth0 name eth0.200 type vlan id 200
ip link add name testbr type bridge
ip link set eth0.100 master testbr
ip link set eth0.200 master testbr
ip link add link testbr mac0 type macvlan
ip link delete dev testbr

This creates an upper/lower tree of (excuse the poor ASCII art):

            /---eth0.100-eth0
mac0-testbr-
            \---eth0.200-eth0

When testbr is deleted, the all_adj_lists are walked, and eth0 is deleted twice from
the mac0 list. Unfortunately, during setup in __netdev_upper_dev_link, only one
reference to eth0 is added, so this results in a panic.

This change adds reference count propagation so things are handled properly.

Matthias Schiffer reported a similar crash in batman-adv:

https://github.com/freifunk-gluon/gluon/issues/680
https://www.open-mesh.org/issues/247

which this patch also seems to resolve.
Signed-off-by: Andrew Collins <acollins@cradlepoint.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Edge-rate:
As system and networking speeds increase, a signal's output transition,
also know as the edge rate or slew rate (V/ns), takes on greater importance
because high-speed signals come with a price. That price is an assortment of
interference problems like ringing on the line, signal overshoot and
undershoot, extended signal settling times, crosstalk noise, transmission
line reflections, false signal detection by the receiving device and
electromagnetic interference (EMI) -- all of which can negate the potential
gains designers are seeking when they try to increase system speeds through
the use of higher performance logic devices. The fact is, faster signaling
edge rates can cause a higher level of electrical noise or other type of
interference that can actually lead to slower line speeds and lower maximum
system frequencies. This parameter allow the board designers to change the
driving strange, and thereby change the EMI behavioral.

Edge-rate parameters (vddmac, edge-slowdown) get from Device Tree.

Tested on Beaglebone Black with VSC 8531 PHY.
Signed-off-by: Raju Lakkaraju <Raju.Lakkaraju@microsemi.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

vmxnet3_reset_work() expects tx queues to be stopped (via
vmxnet3_quiesce_dev -> netif_tx_disable). However, this races with the
netif_wake_queue() call in netif_tx_timeout() such that the driver's
start_xmit routine may be called unexpectedly, triggering one of the BUG_ON
in vmxnet3_map_pkt with a stack trace like this:

RIP: 0010:[<ffffffffa00cf4bc>] vmxnet3_map_pkt+0x3ac/0x4c0 [vmxnet3]
 [<ffffffffa00cf7e0>] vmxnet3_tq_xmit+0x210/0x4e0 [vmxnet3]
 [<ffffffff813ab144>] dev_hard_start_xmit+0x2e4/0x4c0
 [<ffffffff813c956e>] sch_direct_xmit+0x17e/0x1e0
 [<ffffffff813c96a7>] __qdisc_run+0xd7/0x130
 [<ffffffff813a6a7a>] net_tx_action+0x10a/0x200
 [<ffffffff810691df>] __do_softirq+0x11f/0x260
 [<ffffffff81472fdc>] call_softirq+0x1c/0x30
 [<ffffffff81004695>] do_softirq+0x65/0xa0
 [<ffffffff81069b89>] local_bh_enable_ip+0x99/0xa0
 [<ffffffffa031ff36>] destroy_conntrack+0x96/0x110 [nf_conntrack]
 [<ffffffff813d65e2>] nf_conntrack_destroy+0x12/0x20
 [<ffffffff8139c6d5>] skb_release_head_state+0xb5/0xf0
 [<ffffffff8139d299>] skb_release_all+0x9/0x20
 [<ffffffff8139cfe9>] __kfree_skb+0x9/0x90
 [<ffffffffa00d0069>] vmxnet3_quiesce_dev+0x209/0x340 [vmxnet3]
 [<ffffffffa00d020a>] vmxnet3_reset_work+0x6a/0xa0 [vmxnet3]
 [<ffffffff8107d7cc>] process_one_work+0x16c/0x350
 [<ffffffff810804fa>] worker_thread+0x17a/0x410
 [<ffffffff810848c6>] kthread+0x96/0xa0
 [<ffffffff81472ee4>] kernel_thread_helper+0x4/0x10
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Jeff Kirsher says:

====================
100GbE Intel Wired LAN Driver Updates 2016-10-02

This series contains updates to fm10k only.

Jake fixes an issue where PTP applications requesting software timestamps
may complain that the requested mode is not supported, so add a generic
callback for those drivers that have software transmit timestamp support
enabled.  Then provides a trivial cleanup where a code was not wrapped
properly.  Got make sure that code looks good in a 80 character limit.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Although rare, it's possible to hit PCI error early on device
probe, meaning possibly some structs are not entirely initialized,
and some might even be completely uninitialized, leading to NULL
pointer dereference.

The i40e driver currently presents a "bad" behavior if device hits
such early PCI error: firstly, the struct i40e_pf might not be
attached to pci_dev yet, leading to a NULL pointer dereference on
access to pf->state.

Even checking if the struct is NULL and avoiding the access in that
case isn't enough, since the driver cannot recover from PCI error
that early; in our experiments we saw multiple failures on kernel
log, like:

  [549.664] i40e 0007:01:00.1: Initial pf_reset failed: -15
  [549.664] i40e: probe of 0007:01:00.1 failed with error -15
  [...]
  [871.644] i40e 0007:01:00.1: The driver for the device stopped because the
  device firmware failed to init. Try updating your NVM image.
  [871.644] i40e: probe of 0007:01:00.1 failed with error -32
  [...]
  [872.516] i40e 0007:01:00.0: ARQ: Unknown event 0x0000 ignored

Between the first probe failure (error -15) and the second (error -32)
another PCI error happened due to the first bad probe. Also, driver
started to flood console with those ARQ event messages.

This patch will prevent these issues by allowing error recovery
mechanism to remove the failed device from the system instead of
trying to recover from early PCI errors during device probe.

CC: <stable@vger.kernel.org>
Signed-off-by: Guilherme G Piccoli <gpiccoli@linux.vnet.ibm.com>
Acked-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Jeff Kirsher says:

====================
40GbE Intel Wired LAN Driver Updates 2016-10-03

This series contains fixes to i40e only.

Stefan Assmann provides the changes in this series to resolve an issue
where when we run out of MSIx vectors, iWARP gets disabled automatically.
First adds a check for "no vectors left" during MSIx vector allocation
for VMDq, which will prevent more vectors being allocated than available.
Then fixed the MSIx vector redistribution when we reach the hardware limit
for vectors so that additional features like VMDq, iWARP, etc do not get
starved for vectors because the PF is hogging all the resources.  Lastly,
fix the issue for flow director by moving the check for the reaching the
vector limit earlier in the code so that a decision can be made on
disabling flow director.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Yuval Mintz says:

====================
qed*: Add qedr infrastructure support

In the last couple of weeks we've been sending RFCs for the qedr
driver - the RoCE driver for QLogic FastLinQ 4xxxx line of adapters.
Latest RFC can be found at [1].

At Doug's advice [2], we've decided to split the series into two:
 - first part contains the qed backbone that's necessary for all the
configurations relating to the qedr driver, as well as the qede
infrastructure that is used for communication between the qedr and qede.
 - Second part consists of the actual qedr driver and introduces almost
no changes to qed/qede.

This is the first of said two parts. The second half would be sent
later this week.

The only 'oddity' in the devision are the Kconfig options -
As this series introduces both LL2 and QEDR-based logic in qed/qede,
I wanted to add the CONFIG_INFINIBAND_QEDR option here [with default n].
Otherwise, a lot of the code introduced would be dead-code [won't even
be compiled] until qedr is accepted.
As a result I've placed the config option in an odd place - under
qlogic's Kconfig. The second series would then remove that option
and add it in its correct place under the infiniband Kconfig.
[I'm fine with pushing it there to begin with, but I didn't want to
'contaminate' non-qlogic configuration files with half-baked options].

Dave - I don't think you were E-mailed with Doug's suggestion.
I think the notion was to have the two halves accepted side-by-side,
but actually the first has no dependency issues, so it's also
possible to simply take this first to net-next, and push the qedr
into rdma once it's merged. But it's basically up to you and Doug;
We'd align with whatever suits you best.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Add the RoCE-specific LL2 logic [as well as GSI support] over
the 'generic' LL2 interface.
Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add slowpath configuration support for user, dma and memory
regions registration.
Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add support for the slowpath configurations of Queue Pair verbs
which adds, deletes, modifies and queries Queue Pairs.
Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add support for the configurations of the protection domain and
completion queues.
Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This adds the backbone required for the various HW initalizations
which are necessary for the qedr driver - FW notification, resource
initializations, etc.
Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Adds a skeletal implementation of the qede RoCE driver -
The qedr has some dependencies of the state of the underlying base
interface. This adds some logic required with mutual registrations
and the ability to pass updates on 'intresting' events.
Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Other protocols beside the networking driver need the ability
of passing some L2 traffic, usually [although not limited] for the
purpose of some management traffic.
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently if the MSI-X vector limit is reached the sideband flow
director gets disabled. A bit too early to make that decision, as
vectors may get re-distributed. So move the check further back.
Signed-off-by: Stefan Assmann <sassmann@kpanic.de>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

The driver allocates 1 vector per CPU thread and the current hardware
limit for vectors is 129 per PF. On systems with 128 or more threads
this currently means all vectors are used by the PF leaving no room for
additional features like VMDq, iWARP, etc...
The code that should redistribute the vectors in this case is broken and
never triggers. Fixed the code so that it actually triggers if the
hardware limit is reached and adjust the number of queue pairs
accordingly.
Also the number of initially requested iWARP vectors was not properly
saved when the vector limit was reached, and therefore always zero.

Comparison with debug statement.
Before:
i40e 0000:2d:00.0: VMDq disabled, not enough MSI-X vectors
i40e 0000:2d:00.0: IWARP disabled, not enough MSI-X vectors
i40e 00.0 MSI-X vector distribution: PF 128, VMDq 0, FDSB 0, iWARP 0
After:
i40e 0000:2d:00.0: MSI-X vector limit reached, attempting to redistribute vectors
i40e 00.0 MSI-X vector distribution: PF 78, VMDq 8, FDSB 0, iWARP 42
Signed-off-by: Stefan Assmann <sassmann@kpanic.de>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

During MSI-X vector allocation for VMDq, a check for "no vectors left"
was missing, add it. This prevents more vectors to be allocated than
available.
Signed-off-by: Stefan Assmann <sassmann@kpanic.de>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

A call to 'ida_simple_remove()' is missing in the error handling path.

This as been spotted with the following coccinelle script which tries to
detect missing 'ida_simple_remove()' call in error handling paths.

///////////////
@@
expression x;
identifier l;
@@

*   x = ida_simple_get(...);
    ...
    if (...) {
    ...
    }
    ...
    if (...) {
       ...
       goto l;
    }
    ...
*   l: ... when != ida_simple_remove(...);
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In case of error, the function ioremap() returns NULL pointer
not ERR_PTR(). The IS_ERR() test in the return value check
should be replaced with NULL test.

Also add check for return value of platform_get_resource().

Fixes: 54e19bc7 ("net: qcom/emac: do not use devm on internal
phy pdev")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Acked-by: Timur Tabi <timur@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

skb_vlan_pop/push were too generic, trying to support the cases where
skb->data is at mac header, and cases where skb->data is arbitrarily
elsewhere.

Supporting an arbitrary skb->data was complex and bogus:
 - It failed to unwind skb->data to its original location post actual
   pop/push.
   (Also, semantic is not well defined for unwinding: If data was into
    the eth header, need to use same offset from start; But if data was
    at network header or beyond, need to adjust the original offset
    according to the push/pull)
 - It mangled the rcsum post actual push/pop, without taking into account
   that the eth bytes might already have been pulled out of the csum.

Most callers (ovs, bpf) already had their skb->data at mac_header upon
invoking skb_vlan_pop/push.
Last caller that failed to do so (act_vlan) has been recently fixed.

Therefore, to simplify things, no longer support arbitrary skb->data
inputs for skb_vlan_pop/push().

skb->data is expected to be exactly at mac_header; WARN otherwise.
Signed-off-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Pravin Shelar <pshelar@ovn.org>
Cc: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>