net 01/05: fib_rules: rearrange struct fib_rule

commit b8952893d5d86f69c4e499d191b98c6658f64b0f
Author: Patrick McHardy <kaber@trash.net>
Date:   Thu Dec 3 12:05:22 2009 +0100

    net: fib_rules: rearrange struct fib_rule

    The ifname member is only used to resolve interface names and is not needed
    during rule lookups. The target and ctarget members however are used during
    rule lookups and are currently located in a second cacheline.

    Move ifname further to the end to make sure both target and ctarget are
    located in the same cacheline as other members used during rule lookups.

    The layout on 64 bit changes from:

    struct fib_rule {
     ...
            u32                        table;                /*    56     4 */
            u8                         action;               /*    60     1 */

            /* XXX 3 bytes hole, try to pack */

            /* --- cacheline 1 boundary (64 bytes) --- */
            u32                        target;               /*    64     4 */

            /* XXX 4 bytes hole, try to pack */

            struct fib_rule *          ctarget;              /*    72     8 */
            struct rcu_head            rcu;                  /*    80    16 */
            struct net *               fr_net;               /*    96     8 */
    };

    to:

    struct fib_rule {
     ...
            u32                        table;                /*    40     4 */
            u8                         action;               /*    44     1 */

            /* XXX 3 bytes hole, try to pack */

            u32                        target;               /*    48     4 */

            /* XXX 4 bytes hole, try to pack */

            struct fib_rule *          ctarget;              /*    56     8 */
            /* --- cacheline 1 boundary (64 bytes) --- */
            char                       ifname[16];           /*    64    16 */
            struct rcu_head            rcu;                  /*    80    16 */
            struct net *               fr_net;               /*    96     8 */

    };

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

sunhme: fix dma warning in Sun HME

Fix dma-api-checking warnings in Sun HME

Sun HME driver is mapping the first fragment with dma_map_single and subsequent
fragments with dma_map_page. It is unmapping all fragments with dma_unmap_single
and that produces the warning.

This patch changes it so that it unmaps only the first fragment with
dma_unmap_single and subsequent fragments are unmapped with dma_unmap_page.

WARNING: at lib/dma-debug.c:816 check_unmap+0x3ac/0x780()
hme 0000:01:01.1: DMA-API: device driver frees DMA memory with wrong function [device address=0x00000000c1082000] [size=32 bytes] [mapped as page] [unmapped as single]
Modules linked in: nbd sunhme openpromfs sermouse unix
Call Trace:
 [0000000000456910] warn_slowpath_common+0x50/0xa0
 [0000000000571f4c] check_unmap+0x3ac/0x780
 [0000000000572570] debug_dma_unmap_page+0x50/0x60
 [000000001002f5fc] happy_meal_tx+0x11c/0x260 [sunhme]
 [000000001002fc4c] happy_meal_interrupt+0xcc/0xe0 [sunhme]
 [0000000000492d94] handle_fasteoi_irq+0x74/0x100
 [000000000042ac0c] handler_irq+0xcc/0x100
 [0000000000426a54] valid_addr_bitmap_patch+0x14/0x1c0
 [0000000000665de0] _spin_unlock_irqrestore+0x40/0x60
 [0000000000462bb8] mod_timer+0x118/0x1a0
 [00000000005ec254] sk_reset_timer+0x14/0x40
 [0000000000635e4c] tcp_event_new_data_sent+0x8c/0xc0
 [0000000000639374] __tcp_push_pending_frames+0x34/0xc0
---[ end trace 73d5c42c1e9f11c4 ]---
Mapped at:
 [<000000001002f148>] happy_meal_start_xmit+0x308/0x480 [sunhme]
 [<00000000005fc858>] dev_hard_start_xmit+0x318/0x3c0
 [<000000000060fec4>] sch_direct_xmit+0x1a4/0x200
 [<00000000005fced0>] dev_queue_xmit+0x410/0x560
 [<0000000000604a1c>] neigh_resolve_output+0xfc/0x300

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: clear hints to avoid a stale one (nfs only affected?)

Eric Dumazet mentioned in a context of another problem:

"Well, it seems NFS reuses its socket, so maybe we miss some
cleaning as spotted in this old patch"

I've not check under which conditions that actually happens but
if true, we need to make sure we don't accidently leave stale
hints behind when the write queue had to be purged (whether reusing
with NFS can actually happen if purging took place is something I'm
not sure of).

...At least it compiles.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: sysctl_tcp_cookie_size needs to be exported to modules.

Otherwise:

ERROR: "sysctl_tcp_cookie_size" [net/ipv6/ipv6.ko] undefined!
make[1]: *** [__modpost] Error 1

Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: Fix warning on 64-bit.

net/ipv4/tcp_output.c: In function ‘tcp_make_synack’:
net/ipv4/tcp_output.c:2488: warning: cast from pointer to integer of different size

Signed-off-by: David S. Miller <davem@davemloft.net>

net: Teach vlans to cleanup as a pernet subsystem

Take advantage of the fact that an explicit rtnl_kill_links is
unnecessary (and skipping it improves batching), as network namespace
exit calls dellink on all remaining virtual devices, and
rtnl_link_unregister calls dellink on all outstanding devices in that
network namespace.  To do this we need to leave the vlan proc
directories in place until after network device exit time, which is
done by using register_pernet_subsys instead of
register_pernet_device.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

atl1e: Remove non-implementation of ethtool set_msglevel() operation

Unimplemented operations should not silently fail.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

atl1e: Remove redundant definitions of ethtool operations

These functions provide the default behaviour and do not need to be
set in struct ethtool_ops.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

atl1e: Allow TX checksum offload and TSO to be disabled and reenabled

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

TCPCT part 1g: Responder Cookie => Initiator

Parse incoming TCP_COOKIE option(s).

Calculate <SYN,ACK> TCP_COOKIE option.

Send optional <SYN,ACK> data.

This is a significantly revised implementation of an earlier (year-old)
patch that no longer applies cleanly, with permission of the original
author (Adam Langley):

    http://thread.gmane.org/gmane.linux.network/102586

Requires:
   TCPCT part 1a: add request_values parameter for sending SYNACK
   TCPCT part 1b: generate Responder Cookie secret
   TCPCT part 1c: sysctl_tcp_cookie_size, socket option TCP_COOKIE_TRANSACTIONS
   TCPCT part 1d: define TCP cookie option, extend existing struct's
   TCPCT part 1e: implement socket option TCP_COOKIE_TRANSACTIONS
   TCPCT part 1f: Initiator Cookie => Responder

Signed-off-by: William.Allen.Simpson@gmail.com
Signed-off-by: David S. Miller <davem@davemloft.net>

TCPCT part 1f: Initiator Cookie => Responder

Calculate and format <SYN> TCP_COOKIE option.

This is a significantly revised implementation of an earlier (year-old)
patch that no longer applies cleanly, with permission of the original
author (Adam Langley):

    http://thread.gmane.org/gmane.linux.network/102586

Requires:
   TCPCT part 1c: sysctl_tcp_cookie_size, socket option TCP_COOKIE_TRANSACTIONS
   TCPCT part 1d: define TCP cookie option, extend existing struct's

Signed-off-by: William.Allen.Simpson@gmail.com
Signed-off-by: David S. Miller <davem@davemloft.net>

TCPCT part 1e: implement socket option TCP_COOKIE_TRANSACTIONS

Provide per socket control of the TCP cookie option and SYN/SYNACK data.

This is a straightforward re-implementation of an earlier (year-old)
patch that no longer applies cleanly, with permission of the original
author (Adam Langley):

    http://thread.gmane.org/gmane.linux.network/102586

The principle difference is using a TCP option to carry the cookie nonce,
instead of a user configured offset in the data.

Allocations have been rearranged to avoid requiring GFP_ATOMIC.

Requires:
   net: TCP_MSS_DEFAULT, TCP_MSS_DESIRED
   TCPCT part 1c: sysctl_tcp_cookie_size, socket option TCP_COOKIE_TRANSACTIONS
   TCPCT part 1d: define TCP cookie option, extend existing struct's

Signed-off-by: William.Allen.Simpson@gmail.com
Signed-off-by: David S. Miller <davem@davemloft.net>

TCPCT part 1d: define TCP cookie option, extend existing struct's

Data structures are carefully composed to require minimal additions.
For example, the struct tcp_options_received cookie_plus variable fits
between existing 16-bit and 8-bit variables, requiring no additional
space (taking alignment into consideration).  There are no additions to
tcp_request_sock, and only 1 pointer in tcp_sock.

This is a significantly revised implementation of an earlier (year-old)
patch that no longer applies cleanly, with permission of the original
author (Adam Langley):

    http://thread.gmane.org/gmane.linux.network/102586

The principle difference is using a TCP option to carry the cookie nonce,
instead of a user configured offset in the data.  This is more flexible and
less subject to user configuration error.  Such a cookie option has been
suggested for many years, and is also useful without SYN data, allowing
several related concepts to use the same extension option.

    "Re: SYN floods (was: does history repeat itself?)", September 9, 1996.
    http://www.merit.net/mail.archives/nanog/1996-09/msg00235.html

    "Re: what a new TCP header might look like", May 12, 1998.
    ftp://ftp.isi.edu/end2end/end2end-interest-1998.mail

These functions will also be used in subsequent patches that implement
additional features.

Requires:
   TCPCT part 1a: add request_values parameter for sending SYNACK
   TCPCT part 1b: generate Responder Cookie secret
   TCPCT part 1c: sysctl_tcp_cookie_size, socket option TCP_COOKIE_TRANSACTIONS

Signed-off-by: William.Allen.Simpson@gmail.com
Signed-off-by: David S. Miller <davem@davemloft.net>

TCPCT part 1c: sysctl_tcp_cookie_size, socket option TCP_COOKIE_TRANSACTIONS

Define sysctl (tcp_cookie_size) to turn on and off the cookie option
default globally, instead of a compiled configuration option.

Define per socket option (TCP_COOKIE_TRANSACTIONS) for setting constant
data values, retrieving variable cookie values, and other facilities.

Move inline tcp_clear_options() unchanged from net/tcp.h to linux/tcp.h,
near its corresponding struct tcp_options_received (prior to changes).

This is a straightforward re-implementation of an earlier (year-old)
patch that no longer applies cleanly, with permission of the original
author (Adam Langley):

    http://thread.gmane.org/gmane.linux.network/102586

These functions will also be used in subsequent patches that implement
additional features.

Requires:
   net: TCP_MSS_DEFAULT, TCP_MSS_DESIRED

Signed-off-by: William.Allen.Simpson@gmail.com
Signed-off-by: David S. Miller <davem@davemloft.net>

TCPCT part 1b: generate Responder Cookie secret

Define (missing) hash message size for SHA1.

Define hashing size constants specific to TCP cookies.

Add new function: tcp_cookie_generator().

Maintain global secret values for tcp_cookie_generator().

This is a significantly revised implementation of earlier (15-year-old)
Photuris [RFC-2522] code for the KA9Q cooperative multitasking platform.

Linux RCU technique appears to be well-suited to this application, though
neither of the circular queue items are freed.

These functions will also be used in subsequent patches that implement
additional features.

Signed-off-by: William.Allen.Simpson@gmail.com
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

TCPCT part 1a: add request_values parameter for sending SYNACK

Add optional function parameters associated with sending SYNACK.
These parameters are not needed after sending SYNACK, and are not
used for retransmission.  Avoids extending struct tcp_request_sock,
and avoids allocating kernel memory.

Also affects DCCP as it uses common struct request_sock_ops,
but this parameter is currently reserved for future use.

Signed-off-by: William.Allen.Simpson@gmail.com
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'master' of /linux/kernel/git/davem/net-2.6

cnic: Use dma_alloc_coherent().

Replace pci_alloc_consistent() with dma_alloc_coherent() so that
appropriate GFP flags can be used.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Benjamin Li <benli@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cnic: Fix rq_page_table DMA address.

The old code mistakenly zeroed out the upper 32-bit of the DMA address.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Benjamin Li <benli@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cnic: Fix bogus iSCSI MAC address

Fix code to read the proper iSCSI MAC address for bnx2x devices.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Benjamin Li <benli@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cnic: Fix bnx2x ring shutdown.

Need to send a HALT command to the firmware to fully shutdown the bnx2x
rings.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Benjamin Li <benli@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cnic: Fix ring I/O address for bnx2x devices.

Old code was initializing the ring producers using an incorrect I/O
address.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Benjamin Li <benli@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netxen: fix failure cases for fw hang recovery

Handle few corner cases in firmware hang detection and recovery:

o Don't mark device state as READY, till handshake with
  firmware is done.
o During probe, if start_firmware fails, restore reference
  count.
o Don't increment refernce count, if start_firmware fails
  during firmware reset.
o Clear __NX_RESETTING bit, incase of fatal error or tempeature
  reaches critical limit so that pci remove() doesn't poll on
  this bit.

Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netxen : fix debug tools access for NX2031

Driver maps only some part of the MMIO bar 0 in case of
NX2031. This is sufficient for tx and rx by the driver.
When debug tools need to access resources in
unmapped regions, driver needs to map these on the fly
in order to complete read/write request.

Signed-off-by: Narender Kumar <narender.kumar@qlogic.com>
Signed-off-by: Dhananjay Phadke <dhananjay.phadke@qlogic.com>
Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: correct ICH/PCH PHY operations function pointers

Some function pointers for a few PHY operations (for 82578 and 82567) and
were set incorrectly causing functions to be executed that were accessing
incorrect PHY register offsets for that particular device.  This patch also
moves a few PHY-specific functions from ich8lan.c to the more appropriate
phy.c.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: refactor PHY ID detection workaround

The workaround that detects the correct PHY ID when an initial read of the
PHY ID registers returns an invalid one should retry up to ten times with
a small delay between attempts using a single PHY address and then repeat
using the remaining possible PHY addresses.  Do this instead of trying each
possible PHY address repeating that up to 100 times.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: check_polarity function pointers not set for 8257x

The function pointers for 8257x devices are not set.  This is not really a
problem now because there is nothing in the driver that references the
pointers for this particular MAC-family (the appropriate functions are
called directly), but the pointers should be set in case they are used in
the future.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

skbuff: remove skb_dma_map/unmap

The two functions skb_dma_map/unmap are unsafe to use as they cause
problems when packets are cloned and sent to multiple devices while a HW
IOMMU is enabled.  Due to this it is best to remove the code so it is not
used by any other network driver maintainters.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: remove use of skb_dma_map/unmap

Due to the fact that skb_dma_map/unmap do not work correctly when a HW
IOMMU is enabled it has been recommended to go about removing the calls
from the network device drivers.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Matt Carlson <mcarlson@broadcom.com>
CC: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: remove use of skb_dma_map/unmap

Due to the fact that skb_dma_map/unmap do not work correctly when a HW
IOMMU is enabled it has been recommended to go about removing the calls
from the network device drivers.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2: remove skb_dma_map/unmap calls from driver

Due to the fact that skb_dma_map/unmap do not work correctly when a HW
IOMMU is enabled it has been recommended to go about removing the calls
from the network device drivers.

[ Fix bnx2_free_tx_skbs() ring indexing and use NETDEV_TX_OK return
  code in bnx2_start_xmit() after cleaning up DMA mapping errors. -Mchan ]

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igbvf: remove skb_dma_map/unmap call from drivers

This patch removes the skb_dma_map/unmap calls from the igbvf driver due to
the fact that it does not play well with HW IOMMU when combined with
transmitting cloned skbs.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: remove use of skb_dma_map from driver

This change removes skb_dma_map/unmap calls from the igb driver due to the
fact that the call is incompatible with iommu enabled kernels.  In order to
prevent warnings about using the wrong unmap call I have added a
mapped_as_page value to the buffer_info structure to track if the mapped
region is a page or a buffer.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgbe: remove skb_dma_map/unmap calls from driver

This patch removes skb_dma_map/unmap calls from the ixgbe driver due to the
fact that the calls don't work with HW IOMMU enabled systems.  The problem
is that multiple mappings will give different results when HW IOMMU is
enabled and the skb_dma_map/unmap calls only have one location to store
mappings.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Acked-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgb: remove use of skb_dma_map from ixgb

skb_dma_map is incompatible with HW iommu due to the fact that multiple
mappings can result in different results each time.  For this reason it is
best to just remove use of these function calls.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000: remove use of skb_dma_map from e1000 driver

Remove the use of skb_dma_map from the e1000 driver in order to avoid
issues when HW iommu are in use.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: remove use of skb_dma_map from e1000e driver

In testing we have found that skb_dma_map/unmap is incompatible with HW
IOMMU due to the fact that multiple mappings will return different results.
In order to correct this we need to remove skb_dma_map/unmap calls from the
e1000e driver.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip: update the description of rp_filter in ip-sysctl.txt

The commit 27fed4175acf81ddd91d9a4ee2fd298981f60295 (ip: fix logic of
reverse path filter sysctl) has changed the logic of rp_filter.  The
document about rp_filter is out of date. Now, setting
conf/all/rp_filte with 0 can also enable source validation.

Update the document according to the commit.

Signed-off-by: Shan Wei <shanwei@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2: EEH is failing with timeout

bnx2 is failing when a PCI error is detected. The error is the
following:

bnx2: Chip not in correct endian mode
bnx2: fw sync timeout, reset code = 404001d

This error was caused because the way pci_restore_state() is working
after commit 4b77b0a2ba27d64f58f16d8d4d48d8319dda36ff ("PCI: Clear
saved_state after the state has been restored").

Signed-off-by: Breno Leitao<leitao@linux.vnet.ibm.com>
Acked-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Coding style corrections on Sealevel Systems 4021 driver.

Changed braces position on some statements.
Inserted a space between an "if" keyword and a parenthesis.

Signed-off-by: Rudy Matela <rudy.matela@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: compat_sys_recvmmsg user timespec arg can be NULL

We must test if user timespec is non-NULL before copying from userpace,
same as sys_recvmmsg().

Commiter note: changed it so that we have just one branch.

Signed-off-by: Jean-Mickael Guerin <jean-mickael.guerin@6wind.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: compat_mmsghdr must be used in sys_recvmmsg

Both to traverse the entries and to set the msg_len field.

Commiter note: folded two patches and avoided one branch repeating the
compat test.

Signed-off-by: Jean-Mickael Guerin <jean-mickael.guerin@6wind.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sctp: fix sctp_setsockopt_autoclose compile warning

Fix the following warning, when building on 64 bits:

net/sctp/socket.c:2091: warning: large integer implicitly
                        truncated to unsigned type

Signed-off-by: Andrei Pelinescu-Onciul <andrei@iptel.org>
Acked-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ieee802154: merge cleanup

A small cleanup after last net-2.6 merge into net-next-2.6

As we are going to free skb, no need to set skb->skb_iif or skb->dev

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ps3: Add support for gelic link negotiation

Add ethtool_ops.set_settings support to the PS3 gelic network driver.
Allows manual setting of ethernet link speed.

Signed-off-by: Hideyuki Sasaki <xhide@rd.scei.sony.co.jp>
Signed-off-by: Geoff Levand <geoffrey.levand@am.sony.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ps3: Cleanup gelic enums

Cleanup of the gelic driver enumerations:

 o Add some missing commas.
 o Add an Ether port post fix (PS3 currently only supports PORT_0).
 o Add a new enum gelic_lv1_phy to use when interacting with the PHY.

Signed-off-by: Geoff Levand <geoffrey.levand@am.sony.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: fix unnecessary access to hardware to get link, port info

Every time while doing ethtool->get_settings we are accessing
the hardware to get link status and port information.
This is not necessary.  We now use the cached copy for this info
and update it when the link status changes.

From: Suresh R <sureshr@serverengines.com>

Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: Add be_rx_polls counter

Add be_rx_polls to count number of times NAPI called rx poll function.

Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: fix to allow port beacon when device is closed

port identification/beaconing is failing if device open has not been done.
Fixing it.

Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: check for valid ether address

Allow only valid ether addresses to be assigned and used for the interface.

Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: initialize netdev->perm_addr

patch to initialize netdev->perm_addr

Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

be2net: set netdev->vlan_features appropriately

patch to initialize netdev->vlan_features appropriately.

Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

atl1e:disable NETIF_F_TSO6 for hardware limit

For hardware limit to support TSOV6, just disable this feature
Signed-off-by: Jie Yang <jie.yang@atheros.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: Use sizeof struct rather than pointer in e1000_get_eeprom()

Don't use the sizeof the pointer to clear the result

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: PHY type cleanups in e1000e_check_downshift()

Remove the case for 82577 because it does not support the ability to check
for downshift.  Add case for e1000_phy_bm which can do this.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: add debug messages

Add some helpful debug messages.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: comment corrections

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: enable new 82567V-3 device

This new PCI device ID is for a new combination of MAC and PHY both of
which already have supporting code in the driver, just not yet in this
combination.  During validation of the device, an intermittent issue was
discovered with waking it from a suspended state which can be resolved with
the pre-existing workaround to disable gigabit speed prior to suspending.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: ensure the link state is correct for serdes links

This patch ensures that the link state (as reported in
mac->serdes_has_link) will transition to false when autoneg fails to
complete but valid codewords were detected.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: provide family-specific PHY power up/down operations

The different families (80003es2lan, 8257x, ICHx/PCH) supported by the
driver each have their own conditions when the PHY can be powered down.
This patch rewrites the PHY power up/down code to fit with the family-
specific style used in the driver.  All pre-existing calls to power up or
down the PHY remain untouched.  A new call to power down the PHY when
removing the driver when the interface is down replaces the current call
to reset the PHY in order to reduce power consumption.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: guard against buffer overflow in cable length tables

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: provide family-specific functions to manage VLAN filter arrays

The two MAC-families that have VLAN filter table register arrays manage
each a bit differently from one another, so provide family-specific
functions for managing the register arrays and function pointers to access
the appropriate function.  Also make sure attempts to access these
register arrays are not done on parts not supporting that feature.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'master' of /linux/kernel/git/davem/net-2.6

Conflicts:
net/mac80211/ht.c

net: remove [un]register_pernet_gen_... and update the docs.

No that all of the callers have been updated to set fields in
struct pernet_operations, and simplified to let the network
namespace core handle the allocation and freeing of the storage
for them, remove the surpurpflous methods and update the docs
to the new style.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify ipip6 aka sit pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify ip6_tunnel pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify ipip pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify ip_gre pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify phonet pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify pppol2tp pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify pppoe pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify ppp_generic pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify conntrack_proto_gre pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify conntrack_proto_dccp pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify af_key pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify vlan pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify the bond drivers pernet operations.

Take advantage of the new pernet automatic storage management,
and stop using compatibility network namespace functions.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplfy default_device_exit and improve batching.

- Defer dellink to net_cleanup() allowing for batching.
- Fix comment.
- Use for_each_netdev_safe again as dev_change_net_namespace touches
  at most one network device (unlike veth dellink).

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Simplify loopback and improve batching.

Defer calling unregister_netdevice_queue to cleanup_net.  It's simpler
and it allows the loopback device to land in the same batch as other
network devices.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Automatically allocate per namespace data.

To get the full benefit of batched network namespace cleanup netowrk
device deletion needs to be performed by the generic code.  When
using register_pernet_gen_device and freeing the data in exit_net
it is impossible to delay allocation until after exit_net has called
as the device uninit methods are no longer safe.

To correct this, and to simplify working with per network namespace data
I have moved allocation and deletion of per network namespace data into
the network namespace core.  The core now frees the data only after
all of the network namespace exit routines have run.

Now it is only required to set the new fields .id and .size
in the pernet_operations structure if you want network namespace
data to be managed for you automatically.

This makes the current register_pernet_gen_device and
register_pernet_gen_subsys routines unnecessary.  For the moment
I have left them as compatibility wrappers in net_namespace.h
They will be removed once all of the users have been updated.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Batch network namespace destruction.

It is fairly common to kill several network namespaces at once.  Either
because they are nested one inside the other or because they are cooperating
in multiple machine networking experiments.  As the network stack control logic
does not parallelize easily batch up multiple network namespaces existing
together.

To get the full benefit of batching the virtual network devices to be
removed must be all removed in one batch.  For that purpose I have added
a loop after the last network device operations have run that batches
up all remaining network devices and deletes them.

An extra benefit is that the reorganization slightly shrinks the size
of the per network namespace data structures replaceing a work_struct
with a list_head.

In a trivial test with 4K namespaces this change reduced the cost of
a destroying 4K namespaces from 7+ minutes (at 12% cpu) to 44 seconds
(at 60% cpu).  The bulk of that 44s was spent in inet_twsk_purge.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Implement for_each_netdev_reverse.

I will need this shortly to implement network namespace shutdown
batching.  For sanity sake network devices should be removed in
the reverse order they were created in.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: NETDEV_UNREGISTER_PERNET -> NETDEV_UNREGISTER_BATCH

The motivation for an additional notifier in batched netdevice
notification (rt_do_flush) only needs to be called once per batch not
once per namespace.

For further batching improvements I need a guarantee that the
netdevices are unregistered in order allowing me to unregister an all
of the network devices in a network namespace at the same time with
the guarantee that the loopback device is really and truly
unregistered last.

Additionally it appears that we moved the route cache flush after
the final synchronize_net, which seems wrong and there was no
explanation.  So I have restored the original location of the final
synchronize_net.

Cc: Octavian Purdila <opurdila@ixiacom.com>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

MIPS: Wire up recvmmsg syscall

Reported-by: Jean-Mickael Guerin <jean-mickael.guerin@6wind.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip_fragment: also adjust skb->truesize for packets not owned by a socket

When a large packet gets reassembled by ip_defrag(), the head skb
accounts for all the fragments in skb->truesize. If this packet is
refragmented again, skb->truesize is not re-adjusted to reflect only
the head size since its not owned by a socket. If the head fragment
then gets recycled and reused for another received fragment, it might
exceed the defragmentation limits due to its large truesize value.

skb_recycle_check() explicitly checks for linear skbs, so any recycled
skb should reflect its true size in skb->truesize. Change ip_fragment()
to also adjust the truesize value of skbs not owned by a socket.

Reported-and-tested-by: Ben Menchaca <ben@bigfootnetworks.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'master' of /home/davem/src/GIT/linux-2.6/

ipsec: can not add camellia cipher algorithm when using "ip xfrm state" command

can not add camellia cipher algorithm when using "ip xfrm state" command.

Signed-off-by: Li Yewang <lyw@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge git://git./linux/kernel/git/gregkh/staging-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging-2.6:
  Staging: update TODO files
  Staging: hv: Fix some missing author names
  Staging: hv: Fix vmbus event handler bug
  Staging: hv: Fix argument order in incorrect memset invocations in hyperv driver.

Merge git://git./linux/kernel/git/gregkh/usb-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb-2.6:
  USB: Add support for Mobilcom Debitel USB UMTS Surf-Stick to option driver
  USB: work around for EHCI with quirky periodic schedules
  USB: musb: Fix CPPI IRQs not being signaled
  USB: musb: respect usb_request->zero in control requests
  USB: musb: fix ISOC Tx programming for CPPI DMAs
  USB: musb: Remove unwanted message in boot log
  usb: amd5536udc: fixed shared interrupt bug and warning oops
  USB: ftdi_sio: Keep going when write errors are encountered.
  USB: musb_gadget: fix STALL handling
  USB: EHCI: don't send Clear-TT-Buffer following a STALL

Merge git://git./linux/kernel/git/gregkh/tty-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty-2.6:
  tty/of_serial: add missing ns16550a id
  bcm63xx_uart: Fix serial driver compile breakage.
  tty_port: handle the nonblocking open of a dead port corner case

Merge branch 'upstream' of git://ftp.linux-mips.org/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  MIPS: Loongson: Switch from flatmem to sparsemem
  MIPS: Loongson: Disallow 4kB pages
  MIPS: Add missing definition for MADV_HWPOISON.
  MIPS: Fix build error if __xchg() is not getting inlined.
  MIPS: IP22/IP28 Disable early printk to fix boot problems on some systems.

MIPS: Loongson: Switch from flatmem to sparsemem

With flatmem hibernation for Loongson will fail, and there are also some
other problems such as broken files when using NFS or CIFS / Samba.

The config help of sparsemem says:

"This option provides some potential performance benefits, along with
decreased code complexity."

So to avoid the potential problems of FLATMEM, we disable FLATMEM directly
and use SPARSEMEM instead.

Related email thread:

http://groups.google.com/group/loongson-dev/browse_thread/thread/b6b65890ec2b0f24/feb43e5aa7f55d9b?show_docid=feb43e5aa7f55d9b

Reported-by: Tatu Kilappa <tatu.kilappa@gmail.com>
Signed-off-by: Wu Zhangjin <wuzhangjin@gmail.com>
Patchwork: http://patchwork.linux-mips.org/patch/737/
Cc: linux-mips@linux-mips.org
Cc: zhangfx@lemote.com
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

MIPS: Loongson: Disallow 4kB pages

Currently, with PAGE_SIZE_4KB, the kernel for loongson will hang on:

Kernel panic - not syncing: Attempted to kill init!

The possible reason is the cache aliases problem:

Loongson 2F has 64kb, 4 way L1 Cache, the way size is 16kb, which is bigger
then 4kb. so, If using 4kb page size, there is cache aliases problem.
To avoid this kind of problem, extra cache flushing.  The 2nd possible
solution is 16kb page size which avoids cache aliases without the need for
extra cache flushes.  So we disable 4kB pages until the aliasing issue is
solved.

Signed-off-by: Wu Zhangjin <wuzhangjin@gmail.com>
Patchwork: http://patchwork.linux-mips.org/patch/736/
Cc: linux-mips@linux-mips.org
Cc: zhangfx@lemote.com
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

MIPS: Add missing definition for MADV_HWPOISON.

Thanks to Joseph S. Myers for reporting this.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Cc: "Joseph S. Myers" <joseph@codesourcery.com>
Patchwork: http://patchwork.linux-mips.org/patch/723/

MIPS: Fix build error if __xchg() is not getting inlined.

If __xchg() is not getting inlined the outline version of the function
will have a reference to __xchg_called_with_bad_pointer() which does not
exist remaining.  Fixed by using BUILD_BUG_ON() to check for allowable
operand sizes.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Patchwork: http://patchwork.linux-mips.org/patch/705/

MIPS: IP22/IP28 Disable early printk to fix boot problems on some systems.

Some Debian users have reported that the kernel hangs early during boot on
some IP22 systems.  Thomas Bogendoerfer found that this is due to a "bad
interaction between CONFIG_EARLY_PRINTK and overwritten prom memory during
early boot".  Since there's no fix yet, disable CONFIG_EARLY_PRINTK for now.

Signed-off-by: Martin Michlmayr <tbm@cyrius.com>
Cc: linux-mips@linux-mips.org
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Dmitri Vorobiev <dmitri.vorobiev@gmail.com>
Patchwork: http://patchwork.linux-mips.org/patch/702/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

SLOW_WORK: Move slow_work's proc file to debugfs

Move slow_work's debugging proc file to debugfs.

Signed-off-by: David Howells <dhowells@redhat.com>
Requested-and-acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Merge branch 'for-linus' of git://git./linux/kernel/git/mattst88/alpha-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mattst88/alpha-2.6:
  alpha: Fixup last users of irq_chip->typename
  Alpha: Rearrange thread info flags fixing two regressions
  arch/alpha/kernel: Add kmalloc NULL tests
  arch/alpha/kernel/sys_ruffian.c: Use DIV_ROUND_CLOSEST

CacheFiles: Update IMA counters when using dentry_open

When IMA is active, using dentry_open without updating the
IMA counters will result in free/open imbalance errors when
fput is eventually called.

Signed-off-by: Marc Dionne <marc.c.dionne@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

SLOW_WORK: Fix the CONFIG_MODULES=n case

Commits 3d7a641 ("SLOW_WORK: Wait for outstanding work items belonging to a
module to clear") introduced some code to make sure that all of a module's
slow-work items were complete before that module was removed, and commit
3bde31a ("SLOW_WORK: Allow a requeueable work item to sleep till the thread is
needed") further extended that, breaking it in the process if CONFIG_MODULES=n:

    CC      kernel/slow-work.o
  kernel/slow-work.c: In function 'slow_work_execute':
  kernel/slow-work.c:313: error: 'slow_work_thread_processing' undeclared (first use in this function)
  kernel/slow-work.c:313: error: (Each undeclared identifier is reported only once
  kernel/slow-work.c:313: error: for each function it appears in.)
  kernel/slow-work.c: In function 'slow_work_wait_for_items':
  kernel/slow-work.c:950: error: 'slow_work_unreg_sync_lock' undeclared (first use in this function)
  kernel/slow-work.c:951: error: 'slow_work_unreg_wq' undeclared (first use in this function)
  kernel/slow-work.c:961: error: 'slow_work_unreg_work_item' undeclared (first use in this function)
  kernel/slow-work.c:974: error: 'slow_work_unreg_module' undeclared (first use in this function)
  kernel/slow-work.c:977: error: 'slow_work_thread_processing' undeclared (first use in this function)
  make[1]: *** [kernel/slow-work.o] Error 1

Fix this by:

 (1) Extracting the bits of slow_work_execute() that are contingent on
     CONFIG_MODULES, and the bits that should be, into inline functions and
     placing them into the #ifdef'd section that defines the relevant variables
     and adding stubs for moduleless kernels.  This allows the removal of some
     #ifdefs.

 (2) #ifdef'ing out the contents of slow_work_wait_for_items() in moduleless
     kernels.

The four functions related to handling module unloading synchronisation (and
their associated variables) could be offloaded into a separate .c file, but
each function is only used once and three of them are tiny, so doing so would
prevent them from being inlined.

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>