linux-2.6
15 years agoBluetooth: Add CID field to L2CAP socket address structure
Marcel Holtmann [Thu, 12 Feb 2009 04:07:45 +0000 (05:07 +0100)] 
Bluetooth: Add CID field to L2CAP socket address structure

In preparation for L2CAP fixed channel support, the CID value of a
L2CAP connection needs to be accessible via the socket interface. The
CID is the connection identifier and exists as source and destination
value. So extend the L2CAP socket address structure with this field and
change getsockname() and getpeername() to fill it in.

The bind() and connect() functions have been modified to handle L2CAP
socket address structures of variable sizes. This makes them future
proof if additional fields need to be added.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Request L2CAP fixed channel list if available
Marcel Holtmann [Mon, 9 Feb 2009 08:18:02 +0000 (09:18 +0100)] 
Bluetooth: Request L2CAP fixed channel list if available

If the extended features mask indicates support for fixed channels,
request the list of available fixed channels. This also enables the
fixed channel features bit so remote implementations can request
information about it. Currently only the signal channel will be
listed.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Don't enforce authentication for L2CAP PSM 1 and 3
Marcel Holtmann [Mon, 9 Feb 2009 02:55:28 +0000 (03:55 +0100)] 
Bluetooth: Don't enforce authentication for L2CAP PSM 1 and 3

The recommendation for the L2CAP PSM 1 (SDP) is to not use any kind
of authentication or encryption. So don't trigger authentication
for incoming and outgoing SDP connections.

For L2CAP PSM 3 (RFCOMM) there is no clear requirement, but with
Bluetooth 2.1 the initiator is required to enable authentication
and encryption first and this gets enforced. So there is no need
to trigger an additional authentication step. The RFCOMM service
security will make sure that a secure enough link key is present.

When the encryption gets enabled after the SDP connection setup,
then switch the security level from SDP to low security.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Fix double L2CAP connection request
Marcel Holtmann [Fri, 6 Feb 2009 22:56:36 +0000 (23:56 +0100)] 
Bluetooth: Fix double L2CAP connection request

If the remote L2CAP server uses authentication pending stage and
encryption is enabled it can happen that a L2CAP connection request is
sent twice due to a race condition in the connection state machine.

When the remote side indicates any kind of connection pending, then
track this state and skip sending of L2CAP commands for this period.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Fix race condition with L2CAP information request
Marcel Holtmann [Fri, 6 Feb 2009 22:35:19 +0000 (23:35 +0100)] 
Bluetooth: Fix race condition with L2CAP information request

When two L2CAP connections are requested quickly after the ACL link has
been established there exists a window for a race condition where a
connection request is sent before the information response has been
received. Any connection request should only be sent after an exchange
of the extended features mask has been finished.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Set authentication requirements if not available
Marcel Holtmann [Fri, 6 Feb 2009 18:45:36 +0000 (19:45 +0100)] 
Bluetooth: Set authentication requirements if not available

When no authentication requirements are selected, but an outgoing or
incoming connection has requested any kind of security enforcement,
then set these authentication requirements.

This ensures that the userspace always gets informed about the
authentication requirements (if available). Only when no security
enforcement has happened, the kernel will signal invalid requirements.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Use general bonding whenever possible
Marcel Holtmann [Mon, 9 Feb 2009 01:48:38 +0000 (02:48 +0100)] 
Bluetooth: Use general bonding whenever possible

When receiving incoming connection to specific services, always use
general bonding. This ensures that the link key gets stored and can be
used for further authentications.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Add SCO fallback for eSCO connection attempts
Marcel Holtmann [Fri, 6 Feb 2009 08:13:37 +0000 (09:13 +0100)] 
Bluetooth: Add SCO fallback for eSCO connection attempts

When attempting to setup eSCO connections it can happen that some link
manager implementations fail to properly negotiate the eSCO parameters
and thus fail the eSCO setup. Normally the link manager is responsible
for the negotiation of the parameters and actually fallback to SCO if
no agreement can be reached. In cases where the link manager is just too
stupid, then at least try to establish a SCO link if eSCO fails.

For the Bluetooth devices with EDR support this includes handling packet
types of EDR basebands. This is particular tricky since for the EDR the
logic of enabling/disabling one specific packet type is turned around.
This fix contains an extra bitmask to disable eSCO EDR packet when
trying to fallback to a SCO connection.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Don't check encryption for L2CAP raw sockets
Marcel Holtmann [Wed, 4 Feb 2009 20:07:19 +0000 (21:07 +0100)] 
Bluetooth: Don't check encryption for L2CAP raw sockets

For L2CAP sockets with medium and high security requirement a missing
encryption will enforce the closing of the link. For the L2CAP raw
sockets this is not needed, so skip that check.

This fixes a crash when pairing Bluetooth 2.0 (and earlier) devices
since the L2CAP state machine got confused and then locked up the whole
system.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Submit bulk URBs along with interrupt URBs
Marcel Holtmann [Wed, 4 Feb 2009 16:41:38 +0000 (17:41 +0100)] 
Bluetooth: Submit bulk URBs along with interrupt URBs

Submitting the bulk URBs for ACL data transfers only on demand has no
real benefit compared to just submit them when a Bluetooth device gets
opened. So when submitting the interrupt URBs for HCI events, just
submit the bulk URBs, too.

This solves a problem with some Bluetooth USB dongles that has been
reported over the last few month. These devices require that the bulk
URBs are actually present. These devices are really broken, but there
is nothing we can do about it.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: When encryption is dropped, do not send RFCOMM packets
Jaikumar Ganesh [Tue, 3 Feb 2009 02:03:57 +0000 (18:03 -0800)] 
Bluetooth: When encryption is dropped, do not send RFCOMM packets

During a role change with pre-Bluetooth 2.1 devices, the remote side drops
the encryption of the RFCOMM connection. We allow a grace period for the
encryption to be re-established, before dropping the connection. During
this grace period, the RFCOMM_SEC_PENDING flag is set. Check this flag
before sending RFCOMM packets.

Signed-off-by: Jaikumar Ganesh <jaikumar@google.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Eliminate a sparse warning in bt3c driver
Andre Haupt [Mon, 2 Feb 2009 22:45:11 +0000 (14:45 -0800)] 
Bluetooth: Eliminate a sparse warning in bt3c driver

This eliminates a sparse warning that symbol 'stat' shadows an earlier one.

Signed-off-by: Andre Haupt <andre@bitwigglers.org>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Remove CONFIG_DEBUG_LOCK_ALLOC ifdefs
Dave Young [Sat, 31 Jan 2009 05:51:15 +0000 (13:51 +0800)] 
Bluetooth: Remove CONFIG_DEBUG_LOCK_ALLOC ifdefs

Due to lockdep changes, the CONFIG_DEBUG_LOCK_ALLOC ifdef is not needed
now. So just remove it here.

The following commit fixed the !lockdep build warnings:

commit e8f6fbf62de37cbc2e179176ac7010d5f4396b67
Author: Ingo Molnar <mingo@elte.hu>
Date:   Wed Nov 12 01:38:36 2008 +0000

    lockdep: include/linux/lockdep.h - fix warning in net/bluetooth/af_bluetooth.c

Signed-off-by: Dave Young <hidave.darkstar@gmail.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Update version numbers
Marcel Holtmann [Fri, 16 Jan 2009 09:09:50 +0000 (10:09 +0100)] 
Bluetooth: Update version numbers

With the support for the enhanced security model and the support for
deferring connection setup, it is a good idea to increase various
version numbers.

This is purely cosmetic and has no effect on the behavior, but can
be really helpful when debugging problems in different kernel versions.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Restrict application of socket options
Marcel Holtmann [Fri, 16 Jan 2009 09:06:13 +0000 (10:06 +0100)] 
Bluetooth: Restrict application of socket options

The new socket options should only be evaluated for SOL_BLUETOOTH level
and not for every other level. Previously this causes some minor issues
when detecting if a kernel with certain features is available.

Also restrict BT_SECURITY to SOCK_SEQPACKET for L2CAP and SOCK_STREAM for
the RFCOMM protocol.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Disconnect L2CAP connections without encryption
Marcel Holtmann [Thu, 15 Jan 2009 20:58:44 +0000 (21:58 +0100)] 
Bluetooth: Disconnect L2CAP connections without encryption

For L2CAP connections with high security setting, the link will be
immediately dropped when the encryption gets disabled. For L2CAP
connections with medium security there will be grace period where
the remote device has the chance to re-enable encryption. If it
doesn't happen then the link will also be disconnected.

The requirement for the grace period with medium security comes from
Bluetooth 2.0 and earlier devices that require role switching.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Pause RFCOMM TX when encryption drops
Marcel Holtmann [Fri, 16 Jan 2009 07:17:51 +0000 (08:17 +0100)] 
Bluetooth: Pause RFCOMM TX when encryption drops

A role switch with devices following the Bluetooth pre-2.1 standards
or without Encryption Pause and Resume support is not possible if
encryption is enabled. Most newer headsets require the role switch,
but also require that the connection is encrypted.

For connections with a high security mode setting, the link will be
immediately dropped. When the connection uses medium security mode
setting, then a grace period is introduced where the TX is halted and
the remote device gets a change to re-enable encryption after the
role switch. If not re-enabled the link will be dropped.

Based on initial work by Ville Tervo <ville.tervo@nokia.com>

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Replace RFCOMM link mode with security level
Marcel Holtmann [Thu, 15 Jan 2009 20:58:40 +0000 (21:58 +0100)] 
Bluetooth: Replace RFCOMM link mode with security level

Change the RFCOMM internals to use the new security levels and remove
the link mode details.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Replace L2CAP link mode with security level
Marcel Holtmann [Thu, 15 Jan 2009 20:58:38 +0000 (21:58 +0100)] 
Bluetooth: Replace L2CAP link mode with security level

Change the L2CAP internals to use the new security levels and remove
the link mode details.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Add enhanced security model for Simple Pairing
Marcel Holtmann [Thu, 15 Jan 2009 20:58:04 +0000 (21:58 +0100)] 
Bluetooth: Add enhanced security model for Simple Pairing

The current security model is based around the flags AUTH, ENCRYPT and
SECURE. Starting with support for the Bluetooth 2.1 specification this is
no longer sufficient. The different security levels are now defined as
SDP, LOW, MEDIUM and SECURE.

Previously it was possible to set each security independently, but this
actually doesn't make a lot of sense. For Bluetooth the encryption depends
on a previous successful authentication. Also you can only update your
existing link key if you successfully created at least one before. And of
course the update of link keys without having proper encryption in place
is a security issue.

The new security levels from the Bluetooth 2.1 specification are now
used internally. All old settings are mapped to the new values and this
way it ensures that old applications still work. The only limitation
is that it is no longer possible to set authentication without also
enabling encryption. No application should have done this anyway since
this is actually a security issue. Without encryption the integrity of
the authentication can't be guaranteed.

As default for a new L2CAP or RFCOMM connection, the LOW security level
is used. The only exception here are the service discovery sessions on
PSM 1 where SDP level is used. To have similar security strength as with
a Bluetooth 2.0 and before combination key, the MEDIUM level should be
used. This is according to the Bluetooth specification. The MEDIUM level
will not require any kind of man-in-the-middle (MITM) protection. Only
the HIGH security level will require this.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Fix SCO state handling for incoming connections
Marcel Holtmann [Thu, 15 Jan 2009 20:57:03 +0000 (21:57 +0100)] 
Bluetooth: Fix SCO state handling for incoming connections

When the remote device supports only SCO connections, on receipt of
the HCI_EV_CONN_COMPLETE event packet, the connect state is changed to
BT_CONNECTED, but the socket state is not updated. Hence, the connect()
call times out even though the SCO connection has been successfully
established.

Based on a report by Jaikumar Ganesh <jaikumar@google.com>

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Reject incoming SCO connections without listeners
Marcel Holtmann [Thu, 15 Jan 2009 20:57:02 +0000 (21:57 +0100)] 
Bluetooth: Reject incoming SCO connections without listeners

All SCO and eSCO connection are auto-accepted no matter if there is a
corresponding listening socket for them. This patch changes this and
connection requests for SCO and eSCO without any socket are rejected.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Add support for deferring L2CAP connection setup
Marcel Holtmann [Thu, 15 Jan 2009 20:57:00 +0000 (21:57 +0100)] 
Bluetooth: Add support for deferring L2CAP connection setup

In order to decide if listening L2CAP sockets should be accept()ed
the BD_ADDR of the remote device needs to be known. This patch adds
a socket option which defines a timeout for deferring the actual
connection setup.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Add support for deferring RFCOMM connection setup
Marcel Holtmann [Thu, 15 Jan 2009 20:56:48 +0000 (21:56 +0100)] 
Bluetooth: Add support for deferring RFCOMM connection setup

In order to decide if listening RFCOMM sockets should be accept()ed
the BD_ADDR of the remote device needs to be known. This patch adds
a socket option which defines a timeout for deferring the actual
connection setup.

The connection setup is done after reading from the socket for the
first time. Until then writing to the socket returns ENOTCONN.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Add global deferred socket parameter
Marcel Holtmann [Thu, 15 Jan 2009 20:52:16 +0000 (21:52 +0100)] 
Bluetooth: Add global deferred socket parameter

The L2CAP and RFCOMM applications require support for authorization
and the ability of rejecting incoming connection requests. The socket
interface is not really able to support this.

This patch does the ground work for a socket option to defer connection
setup. Setting this option allows calling of accept() and then the
first read() will trigger the final connection setup. Calling close()
would reject the connection.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Preparation for usage of SOL_BLUETOOTH
Marcel Holtmann [Thu, 15 Jan 2009 20:52:14 +0000 (21:52 +0100)] 
Bluetooth: Preparation for usage of SOL_BLUETOOTH

The socket option levels SOL_L2CAP, SOL_RFOMM and SOL_SCO are currently
in use by various Bluetooth applications. Going forward the common
option level SOL_BLUETOOTH should be used. This patch prepares the clean
split of the old and new option levels while keeping everything backward
compatibility.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoBluetooth: Fix issue with return value of rfcomm_sock_sendmsg()
Victor Shcherbatyuk [Thu, 15 Jan 2009 20:52:12 +0000 (21:52 +0100)] 
Bluetooth: Fix issue with return value of rfcomm_sock_sendmsg()

In case of connection failures the rfcomm_sock_sendmsg() should return
an error and not a 0 value.

Signed-off-by: Victor Shcherbatyuk <victor.shcherbatyuk@tomtom.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
15 years agoMerge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
David S. Miller [Wed, 25 Feb 2009 08:02:05 +0000 (00:02 -0800)] 
Merge branch 'master' of /linux/kernel/git/davem/net-2.6

Conflicts:
drivers/net/wireless/orinoco/orinoco.c

15 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
David S. Miller [Wed, 25 Feb 2009 07:52:55 +0000 (23:52 -0800)] 
Merge branch 'master' of git://git./linux/kernel/git/linville/wireless-2.6

15 years agoasix: new device ids
Greg Kroah-Hartman [Wed, 25 Feb 2009 07:52:24 +0000 (23:52 -0800)] 
asix: new device ids

This patch adds two new device ids to the asix driver.

One comes directly from the asix driver on their web site, the other was
reported by Armani Liao as needed for the MSI X320 to get the driver to
work properly for it.

Reported-by: Armani Liao <aliao@novell.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoqlge: Use one path to (re)fill rx buffers.
Ron Mercer [Mon, 23 Feb 2009 10:42:17 +0000 (10:42 +0000)] 
qlge: Use one path to (re)fill rx buffers.

Currently there are two paths for filling rx buffer queues. One is
used during initialization and the other during runtime.  This patch
removes ql_alloc_sbq_buffers() and ql_alloc_lbq_buffers() and replaces
them with a call to the runtime functions ql_update_lbq() and
ql_update_sbq().

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoqlge: Optimize rx buffer refill process.
Ron Mercer [Mon, 23 Feb 2009 10:42:16 +0000 (10:42 +0000)] 
qlge: Optimize rx buffer refill process.

RX Buffers are refilled in chunks of 16 at a time before notifying the
hardware with a register write.  This can cause several writes to take
place in a given napi poll call.  This change causes the write to take place
only once at the end of the call.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoqlge: Change frame route hw semaphore granularity.
Ron Mercer [Mon, 23 Feb 2009 10:42:15 +0000 (10:42 +0000)] 
qlge: Change frame route hw semaphore granularity.

Instead of taking/giving the hw semaphore repeatedly when iterating over
several frame to queue route settings, we have the caller hold it until
all are done.
This reduces PCI bus chatter and possible waits.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoqlge: Increase MAC addr hw sem granularity.
Ron Mercer [Mon, 23 Feb 2009 10:42:14 +0000 (10:42 +0000)] 
qlge: Increase MAC addr hw sem granularity.

Instead of taking/giving the semaphore repeatedly when iterating over
several adderesses, we have the caller hold it until all are done.  This
reduces PCI bus chatter and possible waits.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoqlge: Clean up mac address and frame route settings.
Ron Mercer [Mon, 23 Feb 2009 10:42:13 +0000 (10:42 +0000)] 
qlge: Clean up mac address and frame route settings.

Setting MAC addresses and routing frames to various queues will need to
be done in response to firmware events as well as during initialization.
This change encapsulates the facilities into a single call that can
later me made from other places.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agosit: used time_before for comparing jiffies
Wei Yongjun [Wed, 25 Feb 2009 07:37:19 +0000 (23:37 -0800)] 
sit: used time_before for comparing jiffies

The functions time_before is more robust for comparing
jiffies against other values.

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoipip: used time_before for comparing jiffies
Wei Yongjun [Wed, 25 Feb 2009 07:36:47 +0000 (23:36 -0800)] 
ipip: used time_before for comparing jiffies

The functions time_before is more robust for comparing
jiffies against other values.

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agogre: used time_before for comparing jiffies
Wei Yongjun [Wed, 25 Feb 2009 07:34:48 +0000 (23:34 -0800)] 
gre: used time_before for comparing jiffies

The functions time_before is more robust for comparing
jiffies against other values.

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoipv6: Remove some pointless conditionals before kfree_skb()
Wei Yongjun [Mon, 23 Feb 2009 21:45:33 +0000 (21:45 +0000)] 
ipv6: Remove some pointless conditionals before kfree_skb()

Remove some pointless conditionals before kfree_skb().

The semantic match that finds the problem is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
expression E;
@@
- if (E)
-  kfree_skb(E);
+ kfree_skb(E);
// </smpl>

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetlink: change nlmsg_notify() return value logic
Pablo Neira Ayuso [Wed, 25 Feb 2009 07:18:28 +0000 (23:18 -0800)] 
netlink: change nlmsg_notify() return value logic

This patch changes the return value of nlmsg_notify() as follows:

If NETLINK_BROADCAST_ERROR is set by any of the listeners and
an error in the delivery happened, return the broadcast error;
else if there are no listeners apart from the socket that
requested a change with the echo flag, return the result of the
unicast notification. Thus, with this patch, the unicast
notification is handled in the same way of a broadcast listener
that has set the NETLINK_BROADCAST_ERROR socket flag.

This patch is useful in case that the caller of nlmsg_notify()
wants to know the result of the delivery of a netlink notification
(including the broadcast delivery) and take any action in case
that the delivery failed. For example, ctnetlink can drop packets
if the event delivery failed to provide reliable logging and
state-synchronization at the cost of dropping packets.

This patch also modifies the rtnetlink code to ignore the return
value of rtnl_notify() in all callers. The function rtnl_notify()
(before this patch) returned the error of the unicast notification
which makes rtnl_set_sk_err() reports errors to all listeners. This
is not of any help since the origin of the change (the socket that
requested the echoing) notices the ENOBUFS error if the notification
fails and should resync itself.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agomv643xx_eth: set sane default receive coalescing timeout
Lennert Buytenhek [Tue, 24 Feb 2009 15:42:05 +0000 (15:42 +0000)] 
mv643xx_eth: set sane default receive coalescing timeout

A receive coalescing timeout of 250 usec appears to strike a good
balance between allowing enough received frames to be aggregated for
LRO to do its job and not allowing the connection to stall due to
delaying ACKs to the remote end for too long.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agomv643xx_eth: move a couple of init actions from ->open() to port probe
Lennert Buytenhek [Tue, 24 Feb 2009 15:41:47 +0000 (15:41 +0000)] 
mv643xx_eth: move a couple of init actions from ->open() to port probe

Move the netif_carrier_off() call in ->open() to port probe, so that
ethtool doesn't report the link as being up before we have up'd the
interface.

Move initialisation of the rx/tx coalescing timers from ->open() to
port probe, so that we don't reset the coalescing timers every time
the interface is up'd.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agomv643xx_eth: convert to net_device_ops
Lennert Buytenhek [Tue, 24 Feb 2009 15:41:40 +0000 (15:41 +0000)] 
mv643xx_eth: convert to net_device_ops

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agomv643xx_eth: make LRO unconditional
Lennert Buytenhek [Tue, 24 Feb 2009 15:41:32 +0000 (15:41 +0000)] 
mv643xx_eth: make LRO unconditional

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agotcp_scalable: Update malformed & dead url
Joe Perches [Wed, 25 Feb 2009 00:40:16 +0000 (16:40 -0800)] 
tcp_scalable: Update malformed & dead url

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetxen: refactor pci probe
Dhananjay Phadke [Wed, 25 Feb 2009 00:39:21 +0000 (16:39 -0800)] 
netxen: refactor pci probe

Split pci probe function into smaller logical blocks.

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetxen: cleanup unused code
Dhananjay Phadke [Wed, 25 Feb 2009 00:39:03 +0000 (16:39 -0800)] 
netxen: cleanup unused code

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetxen: advertise wake-on-lan support
Dhananjay Phadke [Wed, 25 Feb 2009 00:38:44 +0000 (16:38 -0800)] 
netxen: advertise wake-on-lan support

Add ethtool wake on lan support.

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetxen: firmware download support
Dhananjay Phadke [Wed, 25 Feb 2009 00:38:22 +0000 (16:38 -0800)] 
netxen: firmware download support

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoixbge: fix bug when using large pages and jumbo frames
Jesse Brandeburg [Wed, 25 Feb 2009 00:37:31 +0000 (16:37 -0800)] 
ixbge: fix bug when using large pages and jumbo frames

it was pointed out on the list that ixgbe was failing when using 64kB pages
and large 16kB MTU.

since with a 64kB PAGE_SIZE MAX_SKB_FRAGS = 3, the way the driver was
configuring page usage was assuming 2kB is half a page, and was only
ever dmaing that much data to a half page.

(16kB - header size) / 2048 = 7 or 8 pages, which would far exceed 3

adjust the driver to account for these large pages, the hardware can
support DMA to up to 16kB for each descriptor.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoixgbe: Move ring features into an enum, allowing easier future maintenance
Shannon Nelson [Wed, 25 Feb 2009 00:36:38 +0000 (16:36 -0800)] 
ixgbe: Move ring features into an enum, allowing easier future maintenance

From: Shannon Nelson <shannon.nelson@intel.com>

The ring_feature member of ixgbe_adapter is statically allocated based on
the supported features of the device.  When a new feature is added, we need
to manually update the static allocation.  This patch makes the feature
list an enum, eliminating the need for multiple updates to the code when
adding a new feature.

Signed-off-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>
15 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/kaber/nf-2.6
David S. Miller [Tue, 24 Feb 2009 21:49:05 +0000 (13:49 -0800)] 
Merge branch 'master' of git://git./linux/kernel/git/kaber/nf-2.6

15 years agonetfilter: xt_recent: fix proc-file addition/removal of IPv4 addresses
Josef Drexler [Tue, 24 Feb 2009 13:53:12 +0000 (14:53 +0100)] 
netfilter: xt_recent: fix proc-file addition/removal of IPv4 addresses

Fix regression introduded by commit 079aa88 (netfilter: xt_recent: IPv6 support):

From http://bugzilla.kernel.org/show_bug.cgi?id=12753:

Problem Description:
An uninitialized buffer causes IPv4 addresses added manually (via the +IP
command to the proc interface) to never match any packets. Similarly, the -IP
command fails to remove IPv4 addresses.

Details:
In the function recent_entry_lookup, the xt_recent module does comparisons of
the entire nf_inet_addr union value, both for IPv4 and IPv6 addresses. For
addresses initialized from actual packets the remaining 12 bytes not occupied
by the IPv4 are zeroed so this works correctly. However when setting the
nf_inet_addr addr variable in the recent_mt_proc_write function, only the IPv4
bytes are initialized and the remaining 12 bytes contain garbage.

Hence addresses added in this way never match any packets, unless these
uninitialized 12 bytes happened to be zero by coincidence. Similarly, addresses
cannot consistently be removed using the proc interface due to mismatch of the
garbage bytes (although it will sometimes work to remove an address that was
added manually).

Reading the /proc/net/xt_recent/ entries hides this problem because this only
uses the first 4 bytes when displaying IPv4 addresses.

Steps to reproduce:
$ iptables -I INPUT -m recent --rcheck -j LOG
$ echo +169.254.156.239 > /proc/net/xt_recent/DEFAULT
$ cat /proc/net/xt_recent/DEFAULT
src=169.254.156.239 ttl: 0 last_seen: 119910 oldest_pkt: 1 119910

[At this point no packets from 169.254.156.239 are being logged.]

$ iptables -I INPUT -s 169.254.156.239 -m recent --set
$ cat /proc/net/xt_recent/DEFAULT
src=169.254.156.239 ttl: 0 last_seen: 119910 oldest_pkt: 1 119910
src=169.254.156.239 ttl: 255 last_seen: 126184 oldest_pkt: 4 125434, 125684, 125934, 126184

[At this point, adding the address via an iptables rule, packets are being
logged correctly.]

$ echo -169.254.156.239 > /proc/net/xt_recent/DEFAULT
$ cat /proc/net/xt_recent/DEFAULT
src=169.254.156.239 ttl: 0 last_seen: 119910 oldest_pkt: 1 119910
src=169.254.156.239 ttl: 255 last_seen: 126992 oldest_pkt: 10 125434, 125684, 125934, 126184, 126434, 126684, 126934, 126991, 126991, 126992
$ echo -169.254.156.239 > /proc/net/xt_recent/DEFAULT
$ cat /proc/net/xt_recent/DEFAULT
src=169.254.156.239 ttl: 0 last_seen: 119910 oldest_pkt: 1 119910
src=169.254.156.239 ttl: 255 last_seen: 126992 oldest_pkt: 10 125434, 125684, 125934, 126184, 126434, 126684, 126934, 126991, 126991, 126992

[Removing the address via /proc interface failed evidently.]

Possible solutions:
- initialize the addr variable in recent_mt_proc_write
- compare only 4 bytes for IPv4 addresses in recent_entry_lookup

Signed-off-by: Patrick McHardy <kaber@trash.net>
15 years agoMerge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
David S. Miller [Tue, 24 Feb 2009 11:51:03 +0000 (03:51 -0800)] 
Merge branch 'master' of /linux/kernel/git/davem/net-2.6

15 years agoMerge branch 'master' of /home/davem/src/GIT/linux-2.6/
David S. Miller [Tue, 24 Feb 2009 11:50:29 +0000 (03:50 -0800)] 
Merge branch 'master' of /home/davem/src/GIT/linux-2.6/

15 years agoMerge branch 'master' of /home/davem/src/GIT/linux-2.6/
David S. Miller [Tue, 24 Feb 2009 11:49:55 +0000 (03:49 -0800)] 
Merge branch 'master' of /home/davem/src/GIT/linux-2.6/

15 years agoDoc: Refer to ip-sysctl.txt for strict vs. loose rp_filter mode
Jesper Dangaard Brouer [Mon, 23 Feb 2009 04:40:43 +0000 (04:40 +0000)] 
Doc: Refer to ip-sysctl.txt for strict vs. loose rp_filter mode

The IP_ADVANCED_ROUTER Kconfig describes the rp_filter
proc option.  Recent changes added a loose mode.
Instead of documenting this change too places, refer to
the document describing it:
 Documentation/networking/ip-sysctl.txt

I'm considering moving the rp_filter description away
from the Kconfig file into ip-sysctl.txt.

Signed-off-by: Jesper Dangaard Brouer <hawk@comx.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoDoc: Cleanup whitespaces in ip-sysctl.txt
Jesper Dangaard Brouer [Mon, 23 Feb 2009 04:39:04 +0000 (04:39 +0000)] 
Doc: Cleanup whitespaces in ip-sysctl.txt

Fix up whitespaces while going though ip-sysctl.txt anyway.

Signed-off-by: Jesper Dangaard Brouer <hawk@comx.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoDoc: Fix typos in ip-sysctl.txt about rp_filter.
Jesper Dangaard Brouer [Mon, 23 Feb 2009 04:37:55 +0000 (04:37 +0000)] 
Doc: Fix typos in ip-sysctl.txt about rp_filter.

First fix a typo in Stephens patch ;-)

Signed-off-by: Jesper Dangaard Brouer <hawk@comx.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetxen: handle pci bar 0 mapping failure
Dhananjay Phadke [Tue, 24 Feb 2009 11:44:23 +0000 (03:44 -0800)] 
netxen: handle pci bar 0 mapping failure

PCI bar 0 is used for memory mapped register access.
If ioremap fails (returns NULL), register access results
in crash.

Use pci_ioremap_bar() instead of ioremap(), the latter
fails on on 32 bit powerpc where pci resource address is
> 32 bits.

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetxen: fix physical port mapping
Dhananjay Phadke [Tue, 24 Feb 2009 11:42:59 +0000 (03:42 -0800)] 
netxen: fix physical port mapping

The PCI function to physical port mapping is valid only for
old firmware. New firmware (4.0.0+) abstracts this.
So driver should never try to access phy using invalid
mapping. The behavior is unpredictable when PCI functions
4-7 are enabled on the same NIC.

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Tue, 24 Feb 2009 04:29:21 +0000 (20:29 -0800)] 
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
  net: amend the fix for SO_BSDCOMPAT gsopt infoleak
  netns: build fix for net_alloc_generic

15 years agoproc: proc_get_inode should de_put when inode already initialized
Krzysztof Sachanowicz [Mon, 23 Feb 2009 21:21:55 +0000 (22:21 +0100)] 
proc: proc_get_inode should de_put when inode already initialized

de_get is called before every proc_get_inode, but corresponding de_put is
called only when dropping last reference to an inode. This might cause
something like
remove_proc_entry: /proc/stats busy, count=14496
to be printed to the syslog.

The fix is to call de_put in case of an already initialized inode in
proc_get_inode.

Signed-off-by: Krzysztof Sachanowicz <analyzer1@gmail.com>
Tested-by: Marcin Pilipczuk <marcin.pilipczuk@gmail.com>
Acked-by: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoi915: suspend/resume interrupt state
Jesse Barnes [Mon, 23 Feb 2009 23:41:09 +0000 (15:41 -0800)] 
i915: suspend/resume interrupt state

In the KMS case, enter/leavevt won't fix up the interrupt handler for
us, so we need to do it at suspend/resume time.  Make sure we don't fail
the resume if the chip is hung either.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoFix an oops in i915_gem_retire_requests()
Karsten Wiese [Mon, 23 Feb 2009 14:07:57 +0000 (15:07 +0100)] 
Fix an oops in i915_gem_retire_requests()

dev_priv->hw_status_page can be NULL, if i915_gem_retire_requests()
is called from i915_gem_busy_ioctl().

Signed-off-by Karsten Wiese <fzu@wemgehoertderstaat.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agonet: amend the fix for SO_BSDCOMPAT gsopt infoleak
Eugene Teo [Mon, 23 Feb 2009 23:38:41 +0000 (15:38 -0800)] 
net: amend the fix for SO_BSDCOMPAT gsopt infoleak

The fix for CVE-2009-0676 (upstream commit df0bca04) is incomplete. Note
that the same problem of leaking kernel memory will reappear if someone
on some architecture uses struct timeval with some internal padding (for
example tv_sec 64-bit and tv_usec 32-bit) --- then, you are going to
leak the padded bytes to userspace.

Signed-off-by: Eugene Teo <eugeneteo@kernel.sg>
Reported-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetns: build fix for net_alloc_generic
Clemens Noss [Mon, 23 Feb 2009 23:37:35 +0000 (15:37 -0800)] 
netns: build fix for net_alloc_generic

net_alloc_generic was defined in #ifdef CONFIG_NET_NS, but used
unconditionally. Move net_alloc_generic out of #ifdef.

Signed-off-by: Clemens Noss <cnoss@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Linus Torvalds [Mon, 23 Feb 2009 22:38:20 +0000 (14:38 -0800)] 
Merge git://git./linux/kernel/git/herbert/crypto-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  crypto: ahash - Fix digest size in /proc/crypto

15 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Mon, 23 Feb 2009 22:36:05 +0000 (14:36 -0800)] 
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
  netns: fix double free at netns creation
  veth : add the set_mac_address capability
  sunlance: Beyond ARRAY_SIZE of ib->btx_ring
  sungem: another error printed one too early
  ISDN: fix sc/shmem printk format warning
  SMSC: timeout reaches -1
  smsc9420: handle magic field of ethtool_eeprom
  sundance: missing parentheses?
  smsc9420: fix another postfixed timeout
  wimax/i2400m: driver loads firmware v1.4 instead of v1.3
  vlan: Update skb->mac_header in __vlan_put_tag().
  cxgb3: Add support for PCI ID 0x35.
  tcp: remove obsoleted comment about different passes
  TG3: &&/|| confusion
  ATM: misplaced parentheses?
  net/mv643xx: don't disable the mib timer too early and lock properly
  net/mv643xx: use GFP_ATOMIC while atomic
  atl1c: Atheros L1C Gigabit Ethernet driver
  net: Kill skb_truesize_check(), it only catches false-positives.
  net: forcedeth: Fix wake-on-lan regression

15 years agortl8187: New USB ID's for RTL8187L
Larry Finger [Tue, 17 Feb 2009 20:31:12 +0000 (14:31 -0600)] 
rtl8187: New USB ID's for RTL8187L

Add new USB ID codes. These come from two postings on forums and
mailing lists, and four are derived from the .inf that accompanies
the latest Realtek Windows driver for the RTL8187L.

Thanks to Viktor Ilijašić <viktor.ilijasic@gmail.com> and Xose Vazquez
Perez <xose.vazquez@gmail.com> for reporting these new ID's.

Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agoath9k: Fix panic upon attach failure
Vasanthakumar Thiagarajan [Mon, 16 Feb 2009 08:25:07 +0000 (13:55 +0530)] 
ath9k: Fix panic upon attach failure

[246916.338046]
[246916.338048] Pid: 29265, comm: insmod Not tainted (2.6.29-rc4-wl #64) 9461DUU
[246916.338051] EIP: 0060:[<c02ca274>] EFLAGS: 00010202 CPU: 0
[246916.338055] EIP is at rollback_registered+0x24/0x220
[246916.338057] EAX: 00000001 EBX: 00000000 ECX: 00000000 EDX: f122e8fc
[246916.338059] ESI: 00000000 EDI: 00000000 EBP: f6595d30 ESP: f6595d1c
[246916.338062]  DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
[246916.338064] Process insmod (pid: 29265, ti=f6594000 task=f7343fe0 task.ti=f6594000)
[246916.338067] Stack:
[246916.338068]  c04a2920 22222222 f6595d48 00000000 f122f080 f6595d48 c02ca489 f122e8fc
[246916.338076]  f122e220 f122f080 f122e220 f6595d5c f8a03156 f122e220 f122f080 f122e220
[246916.338085]  f6595d80 f87359af f122f080 00002000 f874e129 f122f150 f122f080 f6290000
[246916.338094] Call Trace:
[246916.338096]  [<c02ca489>] ? unregister_netdevice+0x19/0x70
[246916.338100]  [<f8a03156>] ? ieee80211_unregister_hw+0x36/0xd0 [mac80211]
[246916.338112]  [<f87359af>] ? ath_detach+0xcf/0x250 [ath9k]
[246916.338127]  [<f8735d9c>] ? ath_attach+0x26c/0x740 [ath9k]
[246916.338139]  [<f873c33a>] ? ath_pci_probe+0x13a/0x310 [ath9k]
[246916.338151]  [<c0233e28>] ? _raw_spin_unlock+0x68/0x80
[246916.338158]  [<c023ab8e>] ? local_pci_probe+0xe/0x10
[246916.338162]  [<c023b8e0>] ? pci_device_probe+0x60/0x80
[246916.338169]  [<c029e042>] ? driver_probe_device+0x82/0x1b0
[246916.338174]  [<c029e1f9>] ? __driver_attach+0x89/0x90
[246916.338180]  [<c029d97b>] ? bus_for_each_dev+0x4b/0x70
[246916.338184]  [<c023b820>] ? pci_device_remove+0x0/0x40
[246916.338190]  [<c029ded9>] ? driver_attach+0x19/0x20
[246916.338193]  [<c029e170>] ? __driver_attach+0x0/0x90
[246916.338197]  [<c029d317>] ? bus_add_driver+0x1b7/0x230
[246916.338203]  [<c023b820>] ? pci_device_remove+0x0/0x40
[246916.338206]  [<c029e399>] ? driver_register+0x69/0x140
[246916.338212]  [<f859d000>] ? ath9k_init+0x0/0x54 [ath9k]
[246916.338221]  [<c023bb4e>] ? __pci_register_driver+0x4e/0x90
[246916.338225]  [<f859d000>] ? ath9k_init+0x0/0x54 [ath9k]
[246916.338232]  [<f859d06b>] ? ath_pci_init+0x17/0x19 [ath9k]
[246916.338238]  [<f859d017>] ? ath9k_init+0x17/0x54 [ath9k]
[246916.338245]  [<c017148e>] ? tracepoint_update_probe_range+0x7e/0xb0
[246916.338249]  [<c010111a>] ? do_one_initcall+0x2a/0x170
[246916.338252]  [<c0149f26>] ? up_read+0x16/0x30
[246916.338256]  [<c014aa9d>] ? __blocking_notifier_call_chain+0x4d/0x60
[246916.338265]  [<c0162b1a>] ? sys_init_module+0x8a/0x1c0
[246916.338269]  [<c022f888>] ? trace_hardirqs_on_thunk+0xc/0x10
[246916.338272]  [<c0103ebf>] ? sysenter_do_call+0x12/0x43
[246916.338276] Code: 8d bc 27 00 00 00 00 55 89 e5 56 89 c6 53 83 ec 0c a1 74 27 4a c0 85 c0 0f 85 4b 01 00 00 e8 04 7d 00 00 85 c0 0f 84 c9 01 00 00 <8b> 86 18 03 00 00 85 c0 0f 84 86 01 00 00 83 e8 01 0f 85 71 01
[246916.338328] EIP: [<c02ca274>] rollback_registered+0x24/0x220 SS:ESP 0068:f6595d1c
[246916.338335] ---[ end trace 76357c56a75ea34e ]---

Signed-off-by: Vasanthakumar Thiagarajan <vasanth@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agoorinoco: do not resgister NULL pm_notifier function
Andrey Borzenkov [Sun, 15 Feb 2009 09:51:18 +0000 (12:51 +0300)] 
orinoco: do not resgister NULL pm_notifier function

With DEBUG_NOTIFIERS it results in

[11330.890966] WARNING: at /home/bor/src/linux-git/kernel/notifier.c:88
notifier_call_chain+0x91/0xa0()
[11330.890977] Hardware name: PORTEGE 4000
[11330.890983] Invalid notifier called! ...

Without DEBUG_NOTIFIERS it most likely crashes on NULL pointer.

Signed-off-by: Andrey Borzenkov <arvidjaar@mail.ru>
Acked-by: David Kilroy <kilroyd@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agoLinux 2.6.29-rc6 v2.6.29-rc6
Linus Torvalds [Mon, 23 Feb 2009 04:19:40 +0000 (20:19 -0800)] 
Linux 2.6.29-rc6

15 years agoacpi/doc: add missing param value
Randy Dunlap [Mon, 23 Feb 2009 01:05:22 +0000 (17:05 -0800)] 
acpi/doc: add missing param value

Add missing parameter value to list of available values
for acpi=<value>.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agonetns: Remove net_alive
Eric W. Biederman [Sun, 22 Feb 2009 08:11:09 +0000 (00:11 -0800)] 
netns: Remove net_alive

It turns out that net_alive is unnecessary, and the original problem
that led to it being added was simply that the icmp code thought
it was a network device and wound up being unable to handle packets
while there were still packets in the network namespace.

Now that icmp and tcp have been fixed to properly register themselves
this problem is no longer present and we have a stronger guarantee
that packets will not arrive in a network namespace then that provided
by net_alive in netif_receive_skb.  So remove net_alive allowing
packet reception run a little faster.

Additionally document the strong reason why network namespace cleanup
is safe so that if something happens again someone else will have
a chance of figuring it out.

Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agotcp: Like icmp use register_pernet_subsys
Eric W. Biederman [Sun, 22 Feb 2009 08:10:18 +0000 (00:10 -0800)] 
tcp: Like icmp use register_pernet_subsys

To remove the possibility of packets flying around when network
devices are being cleaned up use reisger_pernet_subsys instead of
register_pernet_device.

Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Acked-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetns: Fix icmp shutdown.
Eric W. Biederman [Sun, 22 Feb 2009 08:09:14 +0000 (00:09 -0800)] 
netns: Fix icmp shutdown.

Recently I had a kernel panic in icmp_send during a network namespace
cleanup.  There were packets in the arp queue that failed to be sent
and we attempted to generate an ICMP host unreachable message, but
failed because icmp_sk_exit had already been called.

The network devices are removed from a network namespace and their
arp queues are flushed before we do attempt to shutdown subsystems
so this error should have been impossible.

It turns out icmp_init is using register_pernet_device instead
of register_pernet_subsys.  Which resulted in icmp being shut down
while we still had the possibility of packets in flight, making
a nasty NULL pointer deference in interrupt context possible.

Changing this to register_pernet_subsys fixes the problem in
my testing.

Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Acked-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoipv4: Clean whitespaces in net/ipv4/Kconfig.
Jesper Dangaard Brouer [Sun, 22 Feb 2009 08:07:13 +0000 (00:07 -0800)] 
ipv4: Clean whitespaces in net/ipv4/Kconfig.

While going through net/ipv4/Kconfig cleanup whitespaces.

Signed-off-by: Jesper Dangaard Brouer <hawk@comx.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoipv4: Fix rp_filter description in net/ipv4/Kconfig.
Jesper Dangaard Brouer [Sun, 22 Feb 2009 08:06:20 +0000 (00:06 -0800)] 
ipv4: Fix rp_filter description in net/ipv4/Kconfig.

The reverse path filter (rp_filter) will NOT get enabled
when enabling forwarding.  Read the code and tested in
in practice.

Most distributions do enable it in startup scripts.

Signed-off-by: Jesper Dangaard Brouer <hawk@comx.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agosnap: handle registration error and compile warning
Stephen Hemminger [Sun, 22 Feb 2009 08:03:19 +0000 (00:03 -0800)] 
snap: handle registration error and compile warning

If this module can't load, it is almost certainly because something else
is already bound to that SAP. So in that case, return the same error code
as other SAP usage, and fail the module load.

Also fixes a compiler warning about printk of non const.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agollc: fix non-const printk warning
Stephen Hemminger [Sun, 22 Feb 2009 08:02:44 +0000 (00:02 -0800)] 
llc: fix non-const printk warning

Mark some strings as const.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoip: ipip compile warning
Stephen Hemminger [Sun, 22 Feb 2009 08:02:08 +0000 (00:02 -0800)] 
ip: ipip compile warning

Get rid of compile warning about non-const format

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoip: add loose reverse path filtering
Stephen Hemminger [Fri, 20 Feb 2009 08:25:36 +0000 (08:25 +0000)] 
ip: add loose reverse path filtering

Extend existing reverse path filter option to allow strict or loose
filtering. (See http://en.wikipedia.org/wiki/Reverse_path_filtering).

For compatibility with existing usage, the value 1 is chosen for strict mode
and 2 for loose mode.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoRevert "etherh: Get working again."
David S. Miller [Mon, 23 Feb 2009 03:54:16 +0000 (19:54 -0800)] 
Revert "etherh: Get working again."

This reverts commit 7d3d5ec168f1f40e8547d2a1659670ef222a053b.

Russell is going to merge this in via the ARM tree.

Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoMerge branch 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied...
Linus Torvalds [Mon, 23 Feb 2009 00:26:09 +0000 (16:26 -0800)] 
Merge branch 'drm-fixes' of git://git./linux/kernel/git/airlied/drm-2.6

* 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm/i915: Add missing mutex_lock(&dev->struct_mutex)
  drm/i915: fix WC mapping in non-GEM i915 code.
  drm/i915: Fix regression in 95ca9d
  drm/i915: Retire requests from i915_gem_busy_ioctl.
  drm/i915: suspend/resume GEM when KMS is active
  drm/i915: Don't let a device flush to prepare buffers clear new write_domains.
  drm/i915: Cut two args to set_to_gpu_domain that confused this tricky path.

15 years agodrm/i915: Add missing mutex_lock(&dev->struct_mutex)
Pierre Willenbrock [Mon, 23 Feb 2009 00:12:15 +0000 (10:12 +1000)] 
drm/i915: Add missing mutex_lock(&dev->struct_mutex)

there might be a nicer way to fix this but this is the simplest for now.

Signed-off-by: Pierre Willenbrock <pierre@pirsoft.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
15 years agodrm/i915: fix WC mapping in non-GEM i915 code.
Jesse Barnes [Mon, 23 Feb 2009 00:08:21 +0000 (10:08 +1000)] 
drm/i915: fix WC mapping in non-GEM i915 code.

[airlied - taken from mailing list posting]

Signed-off-by: Dave Airlie <airlied@redhat.com>
15 years agodrm/i915: Fix regression in 95ca9d
Chris Wilson [Fri, 20 Feb 2009 17:52:20 +0000 (17:52 +0000)] 
drm/i915: Fix regression in 95ca9d

The object is dereferenced before the NULL check. Oops.

Fixes http://bugs.freedesktop.org/show_bug.cgi?id=20235

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
15 years agodrm/i915: Retire requests from i915_gem_busy_ioctl.
Eric Anholt [Wed, 18 Feb 2009 17:44:56 +0000 (09:44 -0800)] 
drm/i915: Retire requests from i915_gem_busy_ioctl.

This ensures that the user gets the latest information from the hardware
on whether the buffer is busy, potentially reducing the working set of objects
that the user chooses.

Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
15 years agodrm/i915: suspend/resume GEM when KMS is active
Jesse Barnes [Tue, 17 Feb 2009 23:13:31 +0000 (15:13 -0800)] 
drm/i915: suspend/resume GEM when KMS is active

In the KMS case, we need to suspend/resume GEM as well.  So on suspend, make
sure we idle GEM and stop any new rendering from coming in, and on resume,
re-init the framebuffer and clear the suspended flag.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
15 years agodrm/i915: Don't let a device flush to prepare buffers clear new write_domains.
Eric Anholt [Thu, 19 Feb 2009 22:54:51 +0000 (14:54 -0800)] 
drm/i915: Don't let a device flush to prepare buffers clear new write_domains.

The problem was that object_set_to_gpu_domain would set the new write_domains
that are getting set by this batchbuffer, then the accumulated flushes required
for all the objects in preparation for this batchbuffer were posted, and the
brand new write domain would get cleared by the flush being posted.  Instead,
hang on to the new (or old if we're not changing it) value and set it after
the flush is queued.

Results from this noticably included conformance test failures from reads
shortly after writes (where the new write domain had been lost and thus not
flushed and waited on), but is a suspected cause of hangs in some apps when
a write domain is lost on a buffer that gets reused for instruction or
commmand state.

Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
15 years agodrm/i915: Cut two args to set_to_gpu_domain that confused this tricky path.
Eric Anholt [Thu, 19 Feb 2009 22:40:50 +0000 (14:40 -0800)] 
drm/i915: Cut two args to set_to_gpu_domain that confused this tricky path.

While not strictly required, it helped while thinking about the following
change.  This change should be invariant.

Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
15 years agoselinux: Fix the NetLabel glue code for setsockopt()
Paul Moore [Fri, 20 Feb 2009 21:33:02 +0000 (16:33 -0500)] 
selinux: Fix the NetLabel glue code for setsockopt()

At some point we (okay, I) managed to break the ability for users to use the
setsockopt() syscall to set IPv4 options when NetLabel was not active on the
socket in question.  The problem was noticed by someone trying to use the
"-R" (record route) option of ping:

 # ping -R 10.0.0.1
 ping: record route: No message of desired type

The solution is relatively simple, we catch the unlabeled socket case and
clear the error code, allowing the operation to succeed.  Please note that we
still deny users the ability to override IPv4 options on socket's which have
NetLabel labeling active; this is done to ensure the labeling remains intact.

Signed-off-by: Paul Moore <paul.moore@hp.com>
Signed-off-by: James Morris <jmorris@namei.org>
15 years agocipso: Fix documentation comment
Paul Moore [Fri, 20 Feb 2009 21:32:55 +0000 (16:32 -0500)] 
cipso: Fix documentation comment

The CIPSO protocol engine incorrectly stated that the FIPS-188 specification
could be found in the kernel's Documentation directory.  This patch corrects
that by removing the comment and directing users to the FIPS-188 documented
hosted online.  For the sake of completeness I've also included a link to the
CIPSO draft specification on the NetLabel website.

Thanks to Randy Dunlap for spotting the error and letting me know.

Signed-off-by: Paul Moore <paul.moore@hp.com>
Signed-off-by: James Morris <jmorris@namei.org>
15 years agoMerge branch 'core/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux...
Linus Torvalds [Sun, 22 Feb 2009 21:26:30 +0000 (13:26 -0800)] 
Merge branch 'core/urgent' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'core/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  PM: Split up sysdev_[suspend|resume] from device_power_[down|up], fix

15 years agoPM: Split up sysdev_[suspend|resume] from device_power_[down|up], fix
Ingo Molnar [Sun, 22 Feb 2009 21:07:03 +0000 (22:07 +0100)] 
PM: Split up sysdev_[suspend|resume] from device_power_[down|up], fix

Impact: module build fix

Fix:

 ERROR: "sysdev_resume" [arch/x86/kernel/apm.ko] undefined!
 ERROR: "sysdev_suspend" [arch/x86/kernel/apm.ko] undefined!

As these APIs are now used by the APM driver, which can be built
as a module.

Also fix a few extra (and inconsistent) newlines in comment blocks
preceding these functions.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agodocbook: split kernel-api for device-drivers
Randy Dunlap [Sun, 22 Feb 2009 20:15:45 +0000 (12:15 -0800)] 
docbook: split kernel-api for device-drivers

The kernel-api docbook was much larger than any of the others,
so processing it took longer and needed some docbook extras in
some cases, so split it into kernel-api (infrastructure etc.)
and device drivers/device subsystems.  This allows these docbooks
to be generated in parallel.  (This reduced the docbook processing
time on my 4-proc system with make -j4 from about 5min:16sec to
about 2min:01sec.)

The chapters that were moved from kernel-api to device-drivers are:

Driver Basics
Device drivers infrastructure
Parallel Port Devices
Message-based devices
Sound Devices
16x50 UART Driver
Frame Buffer Library
Input Subsystem
Serial Peripheral Interface (SPI)
I2C and SMBus Subsystem

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Acked-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoPM: Split up sysdev_[suspend|resume] from device_power_[down|up]
Rafael J. Wysocki [Sun, 22 Feb 2009 17:38:50 +0000 (18:38 +0100)] 
PM: Split up sysdev_[suspend|resume] from device_power_[down|up]

Move the sysdev_suspend/resume from the callee to the callers, with
no real change in semantics, so that we can rework the disabling of
interrupts during suspend/hibernation.

This is based on an earlier patch from Linus.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agox86: Add IRQF_TIMER to legacy x86 timer interrupt descriptors
Linus Torvalds [Sun, 22 Feb 2009 18:27:49 +0000 (10:27 -0800)] 
x86: Add IRQF_TIMER to legacy x86 timer interrupt descriptors

Right now nobody cares, but the suspend/resume code will eventually want
to suspend device interrupts without suspending the timer, and will
depend on this flag to know.

The modern x86 timer infrastructure uses the local APIC timers and never
shows up as a device interrupt at all, so it isn't affected and doesn't
need any of this.

Cc: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux...
Linus Torvalds [Sun, 22 Feb 2009 17:28:46 +0000 (09:28 -0800)] 
Merge branch 'release' of git://git./linux/kernel/git/lenb/linux-acpi-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux-acpi-2.6:
  ACPI: remove CONFIG_ACPI_SYSTEM
  fujitsu-laptop: Use RFKILL support bitmask from firmware
  x86_64: Fix S3 fail path
  x86_64: acpi/wakeup_64 cleanup
  battery: don't assume we are fully charged when not charging or discharging
  ACPI: EC: Add delay for slow MSI controller