9 years agonetfilter: xtables: change xt_match.checkentry return type
Jan Engelhardt [Fri, 19 Mar 2010 16:16:42 +0000]
netfilter: xtables: change xt_match.checkentry return type

Restore function signatures from bool to int so that we can report
memory allocation failures or similar using -ENOMEM rather than
always having to pass -EINVAL back.

This semantic patch may not be too precise (checking for functions
that use xt_mtchk_param rather than functions referenced by
xt_match.checkentry), but reviewed, it produced the intended result.

// <smpl>
@@
type bool;
identifier check, par;
@@
-bool check
+int check
 (struct xt_mtchk_param *par) { ... }
// </smpl>

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: untangle spaghetti if clauses in checkentry
Jan Engelhardt [Tue, 23 Mar 2010 16:40:13 +0000]
netfilter: xtables: untangle spaghetti if clauses in checkentry

As I'm changing the return values soon, I want to have a clear visual
path.

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: ipvs: use NFPROTO values for NF_HOOK invocation
Jan Engelhardt [Tue, 23 Mar 2010 03:08:46 +0000]
netfilter: ipvs: use NFPROTO values for NF_HOOK invocation

Semantic patch:
// <smpl>
@@
@@
 IP_VS_XMIT(
-PF_INET6,
+NFPROTO_IPV6,
 ...)

@@
@@
 IP_VS_XMIT(
-PF_INET,
+NFPROTO_IPV4,
 ...)
// </smpl>

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: decnet: use NFPROTO values for NF_HOOK invocation
Jan Engelhardt [Tue, 23 Mar 2010 03:09:14 +0000]
netfilter: decnet: use NFPROTO values for NF_HOOK invocation

The semantic patch used was:
// <smpl>
@@
@@
 NF_HOOK(
-PF_DECnet,
+NFPROTO_DECNET,
 ...)
// </smpl>

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: ipv6: use NFPROTO values for NF_HOOK invocation
Jan Engelhardt [Tue, 23 Mar 2010 03:09:07 +0000]
netfilter: ipv6: use NFPROTO values for NF_HOOK invocation

The semantic patch that was used:
// <smpl>
@@
@@
(NF_HOOK
|NF_HOOK_THRESH
|nf_hook
)(
-PF_INET6,
+NFPROTO_IPV6,
 ...)
// </smpl>

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: ipv4: use NFPROTO values for NF_HOOK invocation
Jan Engelhardt [Tue, 23 Mar 2010 03:07:29 +0000]
netfilter: ipv4: use NFPROTO values for NF_HOOK invocation

The semantic patch that was used:
// <smpl>
@@
@@
(NF_HOOK
|NF_HOOK_COND
|nf_hook
)(
-PF_INET,
+NFPROTO_IPV4,
 ...)
// </smpl>

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: bridge: use NFPROTO values for NF_HOOK invocation
Jan Engelhardt [Tue, 23 Mar 2010 03:07:21 +0000]
netfilter: bridge: use NFPROTO values for NF_HOOK invocation

The first argument to NF_HOOK* is an nfproto since quite some time.
Commit v2.6.27-2457-gfdc9314 was the first to practically start using
the new names. Do that now for the remaining NF_HOOK calls.

The semantic patch used was:
// <smpl>
@@
@@
(NF_HOOK
|NF_HOOK_THRESH
)(
-PF_BRIDGE,
+NFPROTO_BRIDGE,
 ...)

@@
@@
 NF_HOOK(
-PF_INET6,
+NFPROTO_IPV6,
 ...)

@@
@@
 NF_HOOK(
-PF_INET,
+NFPROTO_IPV4,
 ...)
// </smpl>

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xt_recent: allow changing ip_list_[ug]id at runtime
Jan Engelhardt [Fri, 19 Mar 2010 20:29:08 +0000]
netfilter: xt_recent: allow changing ip_list_[ug]id at runtime

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: consolidate code into xt_request_find_match
Jan Engelhardt [Fri, 10 Jul 2009 17:27:47 +0000]
netfilter: xtables: consolidate code into xt_request_find_match

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: make use of xt_request_find_target
Jan Engelhardt [Fri, 10 Jul 2009 16:55:11 +0000]
netfilter: xtables: make use of xt_request_find_target

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xt extensions: use pr_<level> (2)
Jan Engelhardt [Fri, 19 Mar 2010 20:08:16 +0000]
netfilter: xt extensions: use pr_<level> (2)

Supplement to 1159683ef48469de71dc26f0ee1a9c30d131cf89.

Downgrade the log level to INFO for most checkentry messages as they
are, IMO, just an extra information to the -EINVAL code that is
returned as part of a parameter "constraint violation". Leave errors
to real errors, such as being unable to create a LED trigger.

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: make use of caller family rather than target family
Jan Engelhardt [Fri, 19 Mar 2010 17:47:51 +0000]
netfilter: xtables: make use of caller family rather than target family

Supplement to aa5fa3185791aac71c9172d4fda3e8729164b5d1.
The semantic patch for this change is:

// <smpl>
@@
struct xt_target_param *par;
@@
-par->target->family
+par->family

@@
struct xt_tgchk_param *par;
@@
-par->target->family
+par->family

@@
struct xt_tgdtor_param *par;
@@
-par->target->family
+par->family
// </smpl>

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: remove unused headers in net/ipv4/netfilter/nf_nat_h323.c
Zhitong Wang [Fri, 19 Mar 2010 15:04:10 +0000]
netfilter: remove unused headers in net/ipv4/netfilter/nf_nat_h323.c

Remove unused headers in net/ipv4/netfilter/nf_nat_h323.c

Signed-off-by: Zhitong Wang <zhitong.wangzt@alibaba-inc.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>

9 years agonetfilter: remove unused headers in net/ipv6/netfilter/ip6t_LOG.c
Zhitong Wang [Fri, 19 Mar 2010 15:01:54 +0000]
netfilter: remove unused headers in net/ipv6/netfilter/ip6t_LOG.c

Remove unused headers in net/ipv6/netfilter/ip6t_LOG.c

Signed-off-by: Zhitong Wang <zhitong.wangzt@alibaba-inc.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>

9 years agonetfilter: xt extensions: use pr_<level>
Jan Engelhardt [Wed, 17 Mar 2010 15:04:40 +0000]
netfilter: xt extensions: use pr_<level>

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: replace custom duprintf with pr_debug
Jan Engelhardt [Thu, 18 Mar 2010 01:22:32 +0000]
netfilter: xtables: replace custom duprintf with pr_debug

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: do not print any messages on ENOMEM
Jan Engelhardt [Wed, 17 Mar 2010 23:27:03 +0000]
netfilter: xtables: do not print any messages on ENOMEM

ENOMEM is a very obvious error code (cf. EINVAL), so I think we do not
really need a warning message. Not to mention that if the allocation
fails, the user is most likely going to get a stack trace from slab
already.

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: reduce holes in struct xt_target
Jan Engelhardt [Thu, 18 Mar 2010 13:02:10 +0000]
netfilter: xtables: reduce holes in struct xt_target

This will save one full padding chunk (8 bytes on x86_64) per target.

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: remove almost-unused xt_match_param.data member
Jan Engelhardt [Thu, 18 Mar 2010 10:03:51 +0000]
netfilter: xtables: remove almost-unused xt_match_param.data member

This member is taking up a "long" per match, yet is only used by one
module out of the roughly 90 modules, ip6t_hbh. ip6t_hbh can be
restructured a little to accomodate for the lack of the .data member.
This variant uses checking the par->match address, which should avoid
having to add two extra functions, including calls, i.e.

(hbh_mt6: call hbhdst_mt6(skb, par, NEXTHDR_OPT),
dst_mt6: call hbhdst_mt6(skb, par, NEXTHDR_DEST))

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: update documentation fields of x_tables.h
Jan Engelhardt [Thu, 18 Mar 2010 09:30:44 +0000]
netfilter: update documentation fields of x_tables.h

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: make use of caller family rather than match family
Jan Engelhardt [Wed, 17 Mar 2010 23:44:52 +0000]
netfilter: xtables: make use of caller family rather than match family

The matches can have .family = NFPROTO_UNSPEC, and though that is not
the case for the touched modules, it seems better to just use the
nfproto from the caller.

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: resort osf kconfig text
Jan Engelhardt [Tue, 16 Mar 2010 19:06:55 +0000]
netfilter: xtables: resort osf kconfig text

Restore alphabetical ordering of the list and put the xt_osf option
into its 'right' place again.

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: limit xt_mac to ethernet devices
Jan Engelhardt [Tue, 16 Mar 2010 20:44:44 +0000]
netfilter: xtables: limit xt_mac to ethernet devices

I do not see a point of allowing the MAC module to work with devices
that don't possibly have one, e.g. various tunnel interfaces such as
tun and sit.

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: clean up xt_mac match routine
Jan Engelhardt [Tue, 16 Mar 2010 20:09:04 +0000]
netfilter: xtables: clean up xt_mac match routine

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: do without explicit XT_ALIGN
Jan Engelhardt [Tue, 9 Mar 2010 22:27:24 +0000]
netfilter: xtables: do without explicit XT_ALIGN

XT_ALIGN is already applied on matchsize/targetsize in x_tables.c,
so it is not strictly needed in the extensions.

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agoMerge branch 'master' of ../nf-2.6
Patrick McHardy [Thu, 18 Mar 2010 11:55:50 +0000]
Merge branch 'master' of ../nf-2.6

9 years agonetfilter: remove unused headers in net/netfilter/nfnetlink.c
Zhitong Wang [Wed, 17 Mar 2010 15:28:25 +0000]
netfilter: remove unused headers in net/netfilter/nfnetlink.c

Remove unused headers in net/netfilter/nfnetlink.c

Signed-off-by: Zhitong Wang <zhitong.wangzt@alibaba-inc.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>

9 years agonetfilter: xt_recent: check for unsupported user space flags
Tim Gardner [Wed, 17 Mar 2010 15:18:56 +0000]
netfilter: xt_recent: check for unsupported user space flags

Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>

9 years agonetfilter: xt_recent: add an entry reaper
Tim Gardner [Tue, 16 Mar 2010 18:53:13 +0000]
netfilter: xt_recent: add an entry reaper

One of the problems with the way xt_recent is implemented is that
there is no efficient way to remove expired entries. Of course,
one can write a rule '-m recent --remove', but you have to know
beforehand which entry to delete. This commit adds reaper
logic which checks the head of the LRU list when a rule
is invoked that has a '--seconds' value and XT_RECENT_REAP set. If an
entry ceases to accumulate time stamps, then it will eventually bubble
to the top of the LRU list where it is then reaped.

Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>

9 years agonetfilter: xt_recent: remove old proc directory
Jan Engelhardt [Mon, 1 Mar 2010 10:55:33 +0000]
netfilter: xt_recent: remove old proc directory

The compat option was introduced in October 2008.

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xt_recent: update description
Jan Engelhardt [Sun, 28 Feb 2010 22:22:35 +0000]
netfilter: xt_recent: update description

It had IPv6 for quite a while already :-)

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: ebt_ip6: add principal maintainer in a MODULE_AUTHOR tag
Jan Engelhardt [Sun, 28 Feb 2010 22:22:04 +0000]
netfilter: ebt_ip6: add principal maintainer in a MODULE_AUTHOR tag

Cc: Kuo-Lang Tseng <kuo-lang.tseng@intel.com>
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: update my email address
Jan Engelhardt [Sun, 28 Feb 2010 22:19:52 +0000]
netfilter: update my email address

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: schedule xt_NOTRACK for removal
Jan Engelhardt [Tue, 16 Mar 2010 17:25:12 +0000]
netfilter: xtables: schedule xt_NOTRACK for removal

It is being superseded by xt_CT (-j CT --notrack).

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: merge xt_CONNMARK into xt_connmark
Jan Engelhardt [Fri, 26 Feb 2010 13:20:32 +0000]
netfilter: xtables: merge xt_CONNMARK into xt_connmark

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: merge xt_MARK into xt_mark
Jan Engelhardt [Sat, 28 Feb 2009 02:23:57 +0000]
netfilter: xtables: merge xt_MARK into xt_mark

Two arguments for combining the two:
- xt_mark is pretty useless without xt_MARK
- the actual code is so small anyway that the kmod metadata and the module
  in its loaded state totally outweighs the combined actual code size.

i586-before:
-rw-r--r-- 1 jengelh users 3821 Feb 10 01:01 xt_MARK.ko
-rw-r--r-- 1 jengelh users 2592 Feb 10 00:04 xt_MARK.o
-rw-r--r-- 1 jengelh users 3274 Feb 10 01:01 xt_mark.ko
-rw-r--r-- 1 jengelh users 2108 Feb 10 00:05 xt_mark.o
   text    data     bss     dec     hex filename
    354     264       0     618     26a xt_MARK.o
    223     176       0     399     18f xt_mark.o
And the runtime size is like 14 KB.

i586-after:
-rw-r--r-- 1 jengelh users 3264 Feb 18 17:28 xt_mark.o

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xtables: add comment markers to Xtables Kconfig
Jan Engelhardt [Fri, 26 Feb 2010 13:14:22 +0000]
netfilter: xtables: add comment markers to Xtables Kconfig

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xt_NFQUEUE: consolidate v4/v6 targets into one
Jan Engelhardt [Fri, 5 Jun 2009 13:22:15 +0000]
netfilter: xt_NFQUEUE: consolidate v4/v6 targets into one

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agonetfilter: xt_CT: par->family is an nfproto
Jan Engelhardt [Wed, 10 Mar 2010 23:38:44 +0000]
netfilter: xt_CT: par->family is an nfproto

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>

9 years agoe1000e: Fix build with CONFIG_PM disabled.
David S. Miller [Wed, 17 Mar 2010 06:36:24 +0000]
e1000e: Fix build with CONFIG_PM disabled.

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

9 years agodrivers/net/e100.c: Use pr_<level> and netif_<level>
Joe Perches [Wed, 17 Mar 2010 04:24:32 +0000]
drivers/net/e100.c: Use pr_<level> and netif_<level>

Convert DPRINTK, commonly used for debugging, to netif_<level>
Remove #define PFX
Use #define pr_fmt
Consistently use no periods for non-sentence logging messages

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoNET: Support clause 45 MDIO commands at the MDIO bus level
Jason Gunthorpe [Tue, 9 Mar 2010 09:17:42 +0000]
NET: Support clause 45 MDIO commands at the MDIO bus level

IEEE 802.3ae clause 45 specifies a somewhat modified MDIO protocol
for use by 10GIGE phys. The main change is a 21 bit address split into
a 5 bit device ID and a 16 bit register offset. The definition is designed
so that normal and extended devices can run on the same MDIO bus.

Extend mdio-bitbang to do the new protocol. At the MDIO bus level the
protocol is requested by or'ing MII_ADDR_C45 into the register offset.

Make phy_read/phy_write/etc pass a full 32 bit register offset.

This does not attempt to make the phy layer support C45 style PHYs, just
to provide the MDIO bus support.

Tested against a Broadcom 10GE phy with ID 0x206034, and several
Broadcom 10/100/1000 Phys in normal mode.

Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoe1000e / PCI / PM: Add basic runtime PM support (rev. 4)
Rafael J. Wysocki [Sun, 14 Mar 2010 14:35:17 +0000]
e1000e / PCI / PM: Add basic runtime PM support (rev. 4)

Use the PCI runtime power management framework to add basic PCI
runtime PM support to the e1000e driver.  Namely, make the driver
suspend the device when the link is off and set it up for generating
a wakeup event after the link has been detected again.  [This
feature is disabled until the user space enables it with the help of
the /sys/devices/.../power/contol device attribute.]

Based on a patch from Matthew Garrett.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agor8169 / PCI / PM: Add simplified runtime PM support (rev. 3)
Rafael J. Wysocki [Sun, 14 Mar 2010 14:33:51 +0000]
r8169 / PCI / PM: Add simplified runtime PM support (rev. 3)

Use the PCI runtime power management framework to add basic PCI
runtime PM support to the r8169 driver.  Namely, make the driver
suspend the device when the link is not present and set it up for
generating a wakeup event after the link has been detected again.
[This feature is disabled until the user space enables it with the
help of the /sys/devices/.../power/contol device attribute.]

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agonet: convert multiple drivers to use netdev_for_each_mc_addr, part7
Jiri Pirko [Mon, 1 Mar 2010 05:09:14 +0000]
net: convert multiple drivers to use netdev_for_each_mc_addr, part7

In mlx4, using char * to store mc address in private structure instead.

Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agodrivers/net/ks*: Use netdev_<level>, netif_<level> and pr_<level>
Joe Perches [Sat, 27 Feb 2010 14:43:51 +0000]
drivers/net/ks*: Use netdev_<level>, netif_<level> and pr_<level>

I'm not sure this is correct.

It changes logging macros from:
dev_<level>(&ks->spidev->dev,
to
netdev_<level>(ks->netdev,

Comments?

Use netdev_<level>
Use netif_<level>
Use pr_<level>
Add #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
Add missing line to message in ks8851_remove
Change kmalloc/memset(,0) to kzalloc
Remove ks_<level> macros
Consolidation code into set_media_state

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

9 years agotipc: Allow retransmission of cloned buffers
Neil Horman [Mon, 15 Mar 2010 07:58:45 +0000]
tipc: Allow retransmission of cloned buffers

Forward port commit
fc477e160af086f6e30c3d4fdf5f5c000d29beb5
from git://tipc.cslab.ericsson.net/pub/git/people/allan/tipc.git

Origional commit message:

Allow retransmission of cloned buffers

This patch fixes an issue with TIPC's message retransmission logic
that prevented retransmission of clone sk_buffs.  Originally intended
as a means of avoiding wasted work in retransmitting messages that
were still on the driver's outbound queue, it also prevented TIPC
from retransmitting messages through other means -- such as the
secondary bearer of the broadcast link, or another interface in a
set of bonded interfaces.  This fix removes existing checks for
cloned sk_buffs that prevented such retransmission.

Origionally-Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agotipc: Increase frequency of load distribution over broadcast link
Neil Horman [Mon, 15 Mar 2010 08:02:24 +0000]
tipc: Increase frequency of load distribution over broadcast link

Forward port commit 29eb572941501c40ac6e62dbc5043bf9ee76ee56
from git://tipc.cslab.ericsson.net/pub/git/people/allan/tipc.git

Origional commit message:
Increase frequency of load distribution over broadcast link

This patch enhances the behavior of TIPC's broadcast link so that it
alternates between redundant bearers (if available) after every
message sent, rather than after every 10 messages.  This change helps
to speed up delivery of retransmitted messages by ensuring that
they are not sent repeatedly over a bearer that is no longer working,
but not yet recognized as failed.

Tested by myself in the latest net-2.6 tree using the tipc sanity test suite

Origionally-signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>

bcast.c |   35 ++++++++++++++---------------------
1 file changed, 14 insertions(+), 21 deletions(-)
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agonet: core: add IFLA_STATS64 support
Jan Engelhardt [Thu, 11 Mar 2010 09:57:29 +0000]
net: core: add IFLA_STATS64 support

`ip -s link` shows interface counters truncated to 32 bit. This is
because interface statistics are transported only in 32-bit quantity
to userspace. This commit adds a new IFLA_STATS64 attribute that
exports them in full 64 bit.

References: http://lkml.indiana.edu/hypermail/linux/kernel/0307.3/0215.html
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agonet: tcp: make veno selectable as default congestion module
Jan Engelhardt [Thu, 11 Mar 2010 09:57:28 +0000]
net: tcp: make veno selectable as default congestion module

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agonet: tcp: make hybla selectable as default congestion module
Jan Engelhardt [Thu, 11 Mar 2010 09:57:27 +0000]
net: tcp: make hybla selectable as default congestion module

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agonet: remove rcu locking from fib_rules_event()
Eric Dumazet [Tue, 9 Mar 2010 20:03:38 +0000]
net: remove rcu locking from fib_rules_event()

We hold RTNL at this point and dont use RCU variants of list traversals,
we dont need rcu_read_lock()/rcu_read_unlock()

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

9 years agobridge: per-cpu packet statistics (v3)
stephen hemminger [Tue, 2 Mar 2010 13:32:09 +0000]
bridge: per-cpu packet statistics (v3)

The shared packet statistics are a potential source of slow down
on bridged traffic. Convert to per-cpu array, but only keep those
statistics which change per-packet.

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

9 years agorps: Receive Packet Steering
Tom Herbert [Tue, 16 Mar 2010 08:03:29 +0000]
rps: Receive Packet Steering

This patch implements software receive side packet steering (RPS).  RPS
distributes the load of received packet processing across multiple CPUs.

Problem statement: Protocol processing done in the NAPI context for received
packets is serialized per device queue and becomes a bottleneck under high
packet load.  This substantially limits pps that can be achieved on a single
queue NIC and provides no scaling with multiple cores.

This solution queues packets early on in the receive path on the backlog queues
of other CPUs.   This allows protocol processing (e.g. IP and TCP) to be
performed on packets in parallel.   For each device (or each receive queue in
a multi-queue device) a mask of CPUs is set to indicate the CPUs that can
process packets. A CPU is selected on a per packet basis by hashing contents
of the packet header (e.g. the TCP or UDP 4-tuple) and using the result to index
into the CPU mask.  The IPI mechanism is used to raise networking receive
softirqs between CPUs.  This effectively emulates in software what a multi-queue
NIC can provide, but is generic requiring no device support.

Many devices now provide a hash over the 4-tuple on a per packet basis
(e.g. the Toeplitz hash).  This patch allow drivers to set the HW reported hash
in an skb field, and that value in turn is used to index into the RPS maps.
Using the HW generated hash can avoid cache misses on the packet when
steering it to a remote CPU.

The CPU mask is set on a per device and per queue basis in the sysfs variable
/sys/class/net/<device>/queues/rx-<n>/rps_cpus.  This is a set of canonical
bit maps for receive queues in the device (numbered by <n>).  If a device
does not support multi-queue, a single variable is used for the device (rx-0).

Generally, we have found this technique increases pps capabilities of a single
queue device with good CPU utilization.  Optimal settings for the CPU mask
seem to depend on architectures and cache hierarcy.  Below are some results
running 500 instances of netperf TCP_RR test with 1 byte req. and resp.
Results show cumulative transaction rate and system CPU utilization.

e1000e on 8 core Intel
   Without RPS: 108K tps at 33% CPU
   With RPS:    311K tps at 64% CPU

forcedeth on 16 core AMD
   Without RPS: 156K tps at 15% CPU
   With RPS:    404K tps at 49% CPU

bnx2x on 16 core AMD
   Without RPS  567K tps at 61% CPU (4 HW RX queues)
   Without RPS  738K tps at 96% CPU (8 HW RX queues)
   With RPS:    854K tps at 76% CPU (4 HW RX queues)

Caveats:
- The benefits of this patch are dependent on architecture and cache hierarchy.
Tuning the masks to get best performance is probably necessary.
- This patch adds overhead in the path for processing a single packet.  In
a lightly loaded server this overhead may eliminate the advantages of
increased parallelism, and possibly cause some relative performance degradation.
We have found that masks that are cache aware (share same caches with
the interrupting CPU) mitigate much of this.
- The RPS masks can be changed dynamically, however whenever the mask is changed
this introduces the possibility of generating out of order packets.  It's
probably best not change the masks too frequently.

Signed-off-by: Tom Herbert <therbert@google.com>

 include/linux/netdevice.h |   32 ++++-
 include/linux/skbuff.h    |    3 +
 net/core/dev.c            |  335 +++++++++++++++++++++++++++++++++++++--------
 net/core/net-sysfs.c      |  225 ++++++++++++++++++++++++++++++-
 net/core/skbuff.c         |    2 +
 5 files changed, 538 insertions(+), 59 deletions(-)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS: Enable per-cpu workqueue threads
Tina Yang [Thu, 11 Mar 2010 13:50:07 +0000]
RDS: Enable per-cpu workqueue threads

Create per-cpu workqueue threads instead of a single
krdsd thread. This is a step towards better scalability.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS: Do not call set_page_dirty() with irqs off
Andy Grover [Thu, 11 Mar 2010 13:50:06 +0000]
RDS: Do not call set_page_dirty() with irqs off

set_page_dirty() unconditionally re-enables interrupts, so
if we call it with irqs off, they will be on after the call,
and that's bad. This patch moves the call after we've re-enabled
interrupts in send_drop_to(), so it's safe.

Also, add BUG_ONs to let us know if we ever do call set_page_dirty
with interrupts off.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS: Properly unmap when getting a remote access error
Sherman Pun [Thu, 11 Mar 2010 13:50:05 +0000]
RDS: Properly unmap when getting a remote access error

If the RDMA op has aborted with a remote access error,
in addition to what we already do (tell userspace it has
completed with an error) also unmap it and put() the rm.

Otherwise, hangs may occur on arches that track maps and
will not exit without proper cleanup.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS: only put sockets that have seen congestion on the poll_waitq
Andy Grover [Thu, 11 Mar 2010 13:50:04 +0000]
RDS: only put sockets that have seen congestion on the poll_waitq

rds_poll_waitq's listeners will be awoken if we receive a congestion
notification. Bad performance may result because *all* polled sockets
contend for this single lock. However, it should not be necessary to
wake pollers when a congestion update arrives if they have never
experienced congestion, and not putting these on the waitq will
hopefully greatly reduce contention.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS: Fix locking in rds_send_drop_to()
Tina Yang [Thu, 11 Mar 2010 13:50:03 +0000]
RDS: Fix locking in rds_send_drop_to()

It seems rds_send_drop_to() called
__rds_rdma_send_complete(rs, rm, RDS_RDMA_CANCELED)
with only rds_sock lock, but not rds_message lock. It raced with
other threads that is attempting to modify the rds_message as well,
such as from within rds_rdma_send_complete().

Signed-off-by: Tina Yang <tina.yang@oracle.com>
Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS: Turn down alarming reconnect messages
Andy Grover [Thu, 11 Mar 2010 13:50:02 +0000]
RDS: Turn down alarming reconnect messages

RDS's error messages when a connection goes down are a little
extreme. A connection may go down, and it will be re-established,
and everything is fine. This patch links these messages through
rdsdebug(), instead of to printk directly.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS: Workaround for in-use MRs on close causing crash
Andy Grover [Thu, 11 Mar 2010 13:50:01 +0000]
RDS: Workaround for in-use MRs on close causing crash

if a machine is shut down without closing sockets properly, and
freeing all MRs, then a BUG_ON will bring it down. This patch
changes these to WARN_ONs -- leaking MRs is not fatal (although
not ideal, and there is more work to do here for a proper fix.)

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS: Fix send locking issue
Tina Yang [Thu, 11 Mar 2010 13:50:00 +0000]
RDS: Fix send locking issue

Fix a deadlock between rds_rdma_send_complete() and
rds_send_remove_from_sock() when rds socket lock and
rds message lock are acquired out-of-order.

Signed-off-by: Tina Yang <Tina.Yang@oracle.com>
Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS: Fix congestion issues for loopback
Andy Grover [Thu, 11 Mar 2010 13:49:59 +0000]
RDS: Fix congestion issues for loopback

We have two kinds of loopback: software (via loop transport)
and hardware (via IB). sw is used for 127.0.0.1, and doesn't
support rdma ops. hw is used for sends to local device IPs,
and supports rdma. Both are used in different cases.

For both of these, when there is a congestion map update, we
want to call rds_cong_map_updated() but not actually send
anything -- since loopback local and foreign congestion maps
point to the same spot, they're already in sync.

The old code never called sw loop's xmit_cong_map(),so
rds_cong_map_updated() wasn't being called for it. sw loop
ports would not work right with the congestion monitor.

Fixing that meant that hw loopback now would send congestion maps
to itself. This is also undesirable (racy), so we check for this
case in the ib-specific xmit code.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS/TCP: Wait to wake thread when write space available
Andy Grover [Thu, 11 Mar 2010 13:49:58 +0000]
RDS/TCP: Wait to wake thread when write space available

Instead of waking the send thread whenever any send space is available,
wait until it is at least half empty. This is modeled on how
sock_def_write_space() does it, and may help to minimize context
switches.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS: update copy_to_user state in tcp transport
Andy Grover [Thu, 11 Mar 2010 13:49:57 +0000]
RDS: update copy_to_user state in tcp transport

Other transports use rds_page_copy_user, which updates our
s_copy_to_user counter. TCP doesn't, so it needs to explicity
call rds_stats_add().

Reported-by: Richard Frank <richard.frank@oracle.com>
Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS: sendmsg() should check sndtimeo, not rcvtimeo
Andy Grover [Thu, 11 Mar 2010 13:49:56 +0000]
RDS: sendmsg() should check sndtimeo, not rcvtimeo

Most likely cut n paste error - sendmsg() was checking sock_rcvtimeo.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoRDS: Do not BUG() on error returned from ib_post_send
Andy Grover [Thu, 11 Mar 2010 13:49:55 +0000]
RDS: Do not BUG() on error returned from ib_post_send

BUGging on a runtime error code should be avoided. This
patch also eliminates all other BUG()s that have no real
reason to exist.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agobridge: Make first arg to deliver_clone const.
David S. Miller [Tue, 16 Mar 2010 21:37:47 +0000]
bridge: Make first arg to deliver_clone const.

Otherwise we get a warning from the call in br_forward().

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

9 years agobridge br_multicast: Don't refer to BR_INPUT_SKB_CB(skb)->mrouters_only without IGMP...
YOSHIFUJI Hideaki / 吉藤英明 [Mon, 15 Mar 2010 21:51:18 +0000]
bridge br_multicast: Don't refer to BR_INPUT_SKB_CB(skb)->mrouters_only without IGMP snooping.

Without CONFIG_BRIDGE_IGMP_SNOOPING,
BR_INPUT_SKB_CB(skb)->mrouters_only is not appropriately
initialized, so we can see garbage.

A clear option to fix this is to set it even without that
config, but we cannot optimize out the branch.

Let's introduce a macro that returns value of mrouters_only
and let it return 0 without CONFIG_BRIDGE_IGMP_SNOOPING.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoroute: Fix caught BUG_ON during rt_secret_rebuild_oneshot()
Vitaliy Gusev [Tue, 16 Mar 2010 01:07:51 +0000]
route: Fix caught BUG_ON during rt_secret_rebuild_oneshot()

route: Fix caught BUG_ON during rt_secret_rebuild_oneshot()

Call rt_secret_rebuild can cause BUG_ON(timer_pending(&net->ipv4.rt_secret_timer)) in
add_timer as there is not any synchronization for call rt_secret_rebuild_oneshot()
for the same net namespace.

Also this issue affects to rt_secret_reschedule().

Thus use mod_timer enstead.

Signed-off-by: Vitaliy Gusev <vgusev@openvz.org>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agobridge br_multicast: Fix skb leakage in error path.
YOSHIFUJI Hideaki / 吉藤英明 [Mon, 15 Mar 2010 19:26:56 +0000]
bridge br_multicast: Fix skb leakage in error path.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agobridge br_multicast: Fix handling of Max Response Code in IGMPv3 message.
YOSHIFUJI Hideaki / 吉藤英明 [Mon, 15 Mar 2010 19:27:00 +0000]
bridge br_multicast: Fix handling of Max Response Code in IGMPv3 message.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoNET: netpoll, fix potential NULL ptr dereference
Jiri Slaby [Tue, 16 Mar 2010 05:29:54 +0000]
NET: netpoll, fix potential NULL ptr dereference

Stanse found that one error path in netpoll_setup dereferences npinfo
even though it is NULL. Avoid that by adding new label and go to that
instead.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Daniel Borkmann <danborkmann@googlemail.com>
Cc: David S. Miller <davem@davemloft.net>
Acked-by: chavey@google.com
Acked-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agotipc: fix lockdep warning on address assignment
Neil Horman [Tue, 16 Mar 2010 08:14:33 +0000]
tipc: fix lockdep warning on address assignment

So in the forward porting of various tipc packages, I was constantly
getting this lockdep warning everytime I used tipc-config to set a network
address for the protocol:

[ INFO: possible circular locking dependency detected ]
2.6.33 #1
tipc-config/1326 is trying to acquire lock:
(ref_table_lock){+.-...}, at: [<ffffffffa0315148>] tipc_ref_discard+0x53/0xd4 [tipc]

but task is already holding lock:
(&(&entry->lock)->rlock#2){+.-...}, at: [<ffffffffa03150d5>] tipc_ref_lock+0x43/0x63 [tipc]

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (&(&entry->lock)->rlock#2){+.-...}:
[<ffffffff8107b508>] __lock_acquire+0xb67/0xd0f
[<ffffffff8107b78c>] lock_acquire+0xdc/0x102
[<ffffffff8145471e>] _raw_spin_lock_bh+0x3b/0x6e
[<ffffffffa03152b1>] tipc_ref_acquire+0xe8/0x11b [tipc]
[<ffffffffa031433f>] tipc_createport_raw+0x78/0x1b9 [tipc]
[<ffffffffa031450b>] tipc_createport+0x8b/0x125 [tipc]
[<ffffffffa030f221>] tipc_subscr_start+0xce/0x126 [tipc]
[<ffffffffa0308fb2>] process_signal_queue+0x47/0x7d [tipc]
[<ffffffff81053e0c>] tasklet_action+0x8c/0xf4
[<ffffffff81054bd8>] __do_softirq+0xf8/0x1cd
[<ffffffff8100aadc>] call_softirq+0x1c/0x30
[<ffffffff810549f4>] _local_bh_enable_ip+0xb8/0xd7
[<ffffffff81054a21>] local_bh_enable_ip+0xe/0x10
[<ffffffff81454d31>] _raw_spin_unlock_bh+0x34/0x39
[<ffffffffa0308eb8>] spin_unlock_bh.clone.0+0x15/0x17 [tipc]
[<ffffffffa0308f47>] tipc_k_signal+0x8d/0xb1 [tipc]
[<ffffffffa0308dd9>] tipc_core_start+0x8a/0xad [tipc]
[<ffffffffa01b1087>] 0xffffffffa01b1087
[<ffffffff8100207d>] do_one_initcall+0x72/0x18a
[<ffffffff810872fb>] sys_init_module+0xd8/0x23a
[<ffffffff81009b42>] system_call_fastpath+0x16/0x1b

-> #0 (ref_table_lock){+.-...}:
[<ffffffff8107b3b2>] __lock_acquire+0xa11/0xd0f
[<ffffffff8107b78c>] lock_acquire+0xdc/0x102
[<ffffffff81454836>] _raw_write_lock_bh+0x3b/0x6e
[<ffffffffa0315148>] tipc_ref_discard+0x53/0xd4 [tipc]
[<ffffffffa03141ee>] tipc_deleteport+0x40/0x119 [tipc]
[<ffffffffa0316e35>] release+0xeb/0x137 [tipc]
[<ffffffff8139dbf4>] sock_release+0x1f/0x6f
[<ffffffff8139dc6b>] sock_close+0x27/0x2b
[<ffffffff811116f6>] __fput+0x12a/0x1df
[<ffffffff811117c5>] fput+0x1a/0x1c
[<ffffffff8110e49b>] filp_close+0x68/0x72
[<ffffffff8110e552>] sys_close+0xad/0xe7
[<ffffffff81009b42>] system_call_fastpath+0x16/0x1b

Finally decided I should fix this.  Its a straightforward inversion,
tipc_ref_acquire takes two locks in this order:
ref_table_lock
entry->lock

while tipc_deleteport takes them in this order:
entry->lock (via tipc_port_lock())
ref_table_lock (via tipc_ref_discard())

when the same entry is referenced, we get the above warning.  The fix is equally
straightforward.  Theres no real relation between the entry->lock and the
ref_table_lock (they just are needed at the same time), so move the entry->lock
aquisition in tipc_ref_acquire down, after we unlock ref_table_lock (this is
safe since the ref_table_lock guards changes to the reference table, and we've
already claimed a slot there.  I've tested the below fix and confirmed that it
clears up the lockdep issue

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agol2tp: Fix UDP socket reference count bugs in the pppol2tp driver
James Chapman [Tue, 16 Mar 2010 06:29:20 +0000]
l2tp: Fix UDP socket reference count bugs in the pppol2tp driver

This patch fixes UDP socket refcnt bugs in the pppol2tp driver.

A bug can cause a kernel stack trace when a tunnel socket is closed.

A way to reproduce the issue is to prepare the UDP socket for L2TP (by
opening a tunnel pppol2tp socket) and then close it before any L2TP
sessions are added to it. The sequence is

Create UDP socket
Create tunnel pppol2tp socket to prepare UDP socket for L2TP
  pppol2tp_connect: session_id=0, peer_session_id=0
L2TP SCCRP control frame received (tunnel_id==0)
  pppol2tp_recv_core: sock_hold()
  pppol2tp_recv_core: sock_put
L2TP ZLB control frame received (tunnel_id=nnn)
  pppol2tp_recv_core: sock_hold()
  pppol2tp_recv_core: sock_put
Close tunnel management socket
  pppol2tp_release: session_id=0, peer_session_id=0
Close UDP socket
  udp_lib_close: BUG

The addition of sock_hold() in pppol2tp_connect() solves the problem.

For data frames, two sock_put() calls were added to plug a refcnt leak
per received data frame. The ref that is grabbed at the top of
pppol2tp_recv_core() must always be released, but this wasn't done for
accepted data frames or data frames discarded because of bad UDP
checksums. This leak meant that any UDP socket that had passed L2TP
data traffic (i.e. L2TP data frames, not just L2TP control frames)
using pppol2tp would not be released by the kernel.

WARNING: at include/net/sock.h:435 udp_lib_unhash+0x117/0x120()
Pid: 1086, comm: openl2tpd Not tainted 2.6.33-rc1 #8
Call Trace:
 [<c119e9b7>] ? udp_lib_unhash+0x117/0x120
 [<c101b871>] ? warn_slowpath_common+0x71/0xd0
 [<c119e9b7>] ? udp_lib_unhash+0x117/0x120
 [<c101b8e3>] ? warn_slowpath_null+0x13/0x20
 [<c119e9b7>] ? udp_lib_unhash+0x117/0x120
 [<c11598a7>] ? sk_common_release+0x17/0x90
 [<c11a5e33>] ? inet_release+0x33/0x60
 [<c11577b0>] ? sock_release+0x10/0x60
 [<c115780f>] ? sock_close+0xf/0x30
 [<c106e542>] ? __fput+0x52/0x150
 [<c106b68e>] ? filp_close+0x3e/0x70
 [<c101d2e2>] ? put_files_struct+0x62/0xb0
 [<c101eaf7>] ? do_exit+0x5e7/0x650
 [<c1081623>] ? mntput_no_expire+0x13/0x70
 [<c106b68e>] ? filp_close+0x3e/0x70
 [<c101eb8a>] ? do_group_exit+0x2a/0x70
 [<c101ebe1>] ? sys_exit_group+0x11/0x20
 [<c10029b0>] ? sysenter_do_call+0x12/0x26

Signed-off-by: James Chapman <jchapman@katalix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agosmsc95xx: wait for PHY to complete reset during init
Steve Glendinning [Tue, 16 Mar 2010 09:03:06 +0000]
smsc95xx: wait for PHY to complete reset during init

This patch ensures the PHY correctly completes its reset before
setting register values.

Signed-off-by: Steve Glendinning <steve.glendinning@smsc.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agol2tp: Fix oops in pppol2tp_xmit
James Chapman [Tue, 16 Mar 2010 06:46:31 +0000]
l2tp: Fix oops in pppol2tp_xmit

When transmitting L2TP frames, we derive the outgoing interface's UDP
checksum hardware assist capabilities from the tunnel dst dev. This
can sometimes be NULL, especially when routing protocols are used and
routing changes occur. This patch just checks for NULL dst or dev
pointers when checking for netdev hardware assist features.

BUG: unable to handle kernel NULL pointer dereference at 0000000c
IP: [<f89d074c>] pppol2tp_xmit+0x341/0x4da [pppol2tp]
*pde = 00000000
Oops: 0000 [#1] SMP
last sysfs file: /sys/class/net/lo/operstate
Modules linked in: pppol2tp pppox ppp_generic slhc ipv6 dummy loop snd_hda_codec_atihdmi snd_hda_intel snd_hda_codec snd_pcm snd_timer snd soundcore snd_page_alloc evdev psmouse serio_raw processor button i2c_piix4 i2c_core ati_agp agpgart pcspkr ext3 jbd mbcache sd_mod ide_pci_generic atiixp ide_core ahci ata_generic floppy ehci_hcd ohci_hcd libata e1000e scsi_mod usbcore nls_base thermal fan thermal_sys [last unloaded: scsi_wait_scan]

Pid: 0, comm: swapper Not tainted (2.6.32.8 #1)
EIP: 0060:[<f89d074c>] EFLAGS: 00010297 CPU: 3
EIP is at pppol2tp_xmit+0x341/0x4da [pppol2tp]
EAX: 00000000 EBX: f64d1680 ECX: 000005b9 EDX: 00000000
ESI: f6b91850 EDI: f64d16ac EBP: f6a0c4c0 ESP: f70a9cac
 DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
Process swapper (pid: 0, ti=f70a8000 task=f70a31c0 task.ti=f70a8000)
Stack:
 000005a9 000005b9 f734c400 f66652c0 f7352e00 f67dc800 00000000 f6b91800
<0> 000005a3 f70ef6c4 f67dcda9 000005a3 f89b192e 00000246 000005a3 f64d1680
<0> f63633e0 f6363320 f64d1680 f65a7320 f65a7364 f65856c0 f64d1680 f679f02f
Call Trace:
 [<f89b192e>] ? ppp_push+0x459/0x50e [ppp_generic]
 [<f89b217f>] ? ppp_xmit_process+0x3b6/0x430 [ppp_generic]
 [<f89b2306>] ? ppp_start_xmit+0x10d/0x120 [ppp_generic]
 [<c11c15cb>] ? dev_hard_start_xmit+0x21f/0x2b2
 [<c11d0947>] ? sch_direct_xmit+0x48/0x10e
 [<c11c19a0>] ? dev_queue_xmit+0x263/0x3a6
 [<c11e2a9f>] ? ip_finish_output+0x1f7/0x221
 [<c11df682>] ? ip_forward_finish+0x2e/0x30
 [<c11de645>] ? ip_rcv_finish+0x295/0x2a9
 [<c11c0b19>] ? netif_receive_skb+0x3e9/0x404
 [<f814b791>] ? e1000_clean_rx_irq+0x253/0x2fc [e1000e]
 [<f814cb7a>] ? e1000_clean+0x63/0x1fc [e1000e]
 [<c1047eff>] ? sched_clock_local+0x15/0x11b
 [<c11c1095>] ? net_rx_action+0x96/0x195
 [<c1035750>] ? __do_softirq+0xaa/0x151
 [<c1035828>] ? do_softirq+0x31/0x3c
 [<c10358fe>] ? irq_exit+0x26/0x58
 [<c1004b21>] ? do_IRQ+0x78/0x89
 [<c1003729>] ? common_interrupt+0x29/0x30
 [<c101ac28>] ? native_safe_halt+0x2/0x3
 [<c1008c54>] ? default_idle+0x55/0x75
 [<c1009045>] ? c1e_idle+0xd2/0xd5
 [<c100233c>] ? cpu_idle+0x46/0x62
Code: 8d 45 08 f0 ff 45 08 89 6b 08 c7 43 68 7e fb 9c f8 8a 45 24 83 e0 0c 3c 04 75 09 80 63 64 f3 e9 b4 00 00 00 8b 43 18 8b 4c 24 04 <8b> 40 0c 8d 79 11 f6 40 44 0e 8a 43 64 75 51 6a 00 8b 4c 24 08
EIP: [<f89d074c>] pppol2tp_xmit+0x341/0x4da [pppol2tp] SS:ESP 0068:f70a9cac
CR2: 000000000000000c

Signed-off-by: James Chapman <jchapman@katalix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agosmsc75xx: SMSC LAN75xx USB gigabit ethernet adapter driver
Steve Glendinning [Tue, 16 Mar 2010 08:46:46 +0000]
smsc75xx: SMSC LAN75xx USB gigabit ethernet adapter driver

This patch adds a driver for SMSC's LAN7500 family of USB 2.0
to gigabit ethernet adapters.  It's loosely based on the smsc95xx
driver but the device registers for LAN7500 are completely different.

Signed-off-by: Steve Glendinning <steve.glendinning@smsc.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agone: Do not use slashes in irq name string
Atsushi Nemoto [Tue, 16 Mar 2010 05:27:40 +0000]
ne: Do not use slashes in irq name string

This patch fixes following warning introduced by commit
12bac0d9f4dbf3445a0319beee848d15fa32775e ("proc: warn on non-existing
proc entries"):

WARNING: at /work/mips-linux/make/linux/fs/proc/generic.c:316 __xlate_proc_name+0xe0/0xe8()
name 'RBHMA4X00/RTL8019'

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoNET: ksz884x, fix lock imbalance
Jiri Slaby [Tue, 16 Mar 2010 04:53:50 +0000]
NET: ksz884x, fix lock imbalance

Stanse found that one error path (when alloc_skb fails) in netdev_tx
omits to unlock hw_priv->hwlock. Fix that.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Tristram Ha <Tristram.Ha@micrel.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agogigaset: correct range checking off by one error
Tilman Schmidt [Tue, 16 Mar 2010 07:04:01 +0000]
gigaset: correct range checking off by one error

Correct a potential array overrun due to an off by one error in the
range check on the CAPI CONNECT_REQ CIPValue parameter.
Found and reported by Dan Carpenter using smatch.

Impact: bugfix
Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agobridge: Fix br_forward crash in promiscuous mode
Michael Braun [Tue, 16 Mar 2010 07:26:22 +0000]
bridge: Fix br_forward crash in promiscuous mode

From: Michael Braun <michael-dev@fami-braun.de>

bridge: Fix br_forward crash in promiscuous mode

It's a linux-next kernel from 2010-03-12 on an x86 system and it
OOPs in the bridge module in br_pass_frame_up (called by
br_handle_frame_finish) because brdev cannot be dereferenced (its set to
a non-null value).

Adding some BUG_ON statements revealed that
 BR_INPUT_SKB_CB(skb)->brdev == br-dev
(as set in br_handle_frame_finish first)
only holds until br_forward is called.
The next call to br_pass_frame_up then fails.

Digging deeper it seems that br_forward either frees the skb or passes
it to NF_HOOK which will in turn take care of freeing the skb. The
same is holds for br_pass_frame_ip. So it seems as if two independent
skb allocations are required. As far as I can see, commit
b33084be192ee1e347d98bb5c9e38a53d98d35e2 ("bridge: Avoid unnecessary
clone on forward path") removed skb duplication and so likely causes
this crash. This crash does not happen on 2.6.33.

I've therefore modified br_forward the same way br_flood has been
modified so that the skb is not freed if skb0 is going to be used
and I can confirm that the attached patch resolves the issue for me.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agobridge: Move NULL mdb check into br_mdb_ip_get
Herbert Xu [Tue, 16 Mar 2010 03:38:25 +0000]
bridge: Move NULL mdb check into br_mdb_ip_get

Since all callers of br_mdb_ip_get need to check whether the
hash table is NULL, this patch moves the check into the function.

This fixes the two callers (query/leave handler) that didn't
check it.

Reported-by: Michael Braun <michael-dev@fami-braun.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoISDN: Add PCI ID for HFC-2S/4S Beronet Card PCIe
Lars Ellenberg [Tue, 16 Mar 2010 02:09:28 +0000]
ISDN: Add PCI ID for HFC-2S/4S Beronet Card PCIe

A few subdevice IDs seem to have been dropped when hfc_multi was
included upstream, just compare the list at
http://www.openvox.cn/viewvc/misdn/trunk/hfc_multi.c?revision=75&view=annotate#l175
with the IDs in drivers/isdn/hardware/mISDN/hfcmulti.c

Added PCIe 2 Port card and LED settings (same as PCI)
Do not use <linux/pci_ids.h> /KKe

Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Karsten Keil <keil@b1-systems.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoMerge branch 'master' of /home/davem/src/GIT/linux-2.6/
David S. Miller [Mon, 15 Mar 2010 23:23:54 +0000]
Merge branch 'master' of /home/davem/src/GIT/linux-2.6/

9 years agonet-2.6 [Bug-Fix][dccp]: fix oops caused after failed initialisation
Gerrit Renker [Sun, 14 Mar 2010 20:13:19 +0000]
net-2.6 [Bug-Fix][dccp]: fix oops caused after failed initialisation

dccp: fix panic caused by failed initialisation

This fixes a kernel panic reported thanks to Andre Noll:

if DCCP is compiled into the kernel and any out of the initialisation
steps in net/dccp/proto.c:dccp_init() fail, a subsequent attempt to create
a SOCK_DCCP socket will panic, since inet{,6}_create() are not prevented
from creating DCCP sockets.

This patch fixes the problem by propagating a failure in dccp_init() to
dccp_v{4,6}_init_net(), and from there to dccp_v{4,6}_init(), so that the
DCCP protocol is not made available if its initialisation fails.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agomyri: remove dead code
Dan Carpenter [Sun, 14 Mar 2010 22:24:08 +0000]
myri: remove dead code

We can never reach the return statement.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agogigaset: prune use of tty_buffer_request_room
Tilman Schmidt [Sun, 14 Mar 2010 12:58:05 +0000]
gigaset: prune use of tty_buffer_request_room

Calling tty_buffer_request_room() before tty_insert_flip_string()
is unnecessary, costs CPU and for big buffers can mess up the
multi-page allocation avoidance.

Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Acked-by: Karsten Keil <keil@b1-systems.de>
CC: Alan Cox <alan@lxorguk.ukuu.org.uk>, stable@kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agogigaset: correct clearing of at_state strings on RING
Tilman Schmidt [Sun, 14 Mar 2010 12:58:05 +0000]
gigaset: correct clearing of at_state strings on RING

In RING handling, clear the table of received parameter strings in
a loop like everywhere else, instead of by enumeration which had
already gotten out of sync.

Impact: minor bugfix
Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Acked-by: Karsten Keil <keil@b1-systems.de>
CC: stable@kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agogigaset: avoid registering CAPI driver more than once
Tilman Schmidt [Sun, 14 Mar 2010 12:58:05 +0000]
gigaset: avoid registering CAPI driver more than once

Registering/unregistering the Gigaset CAPI driver when a device is
connected/disconnected causes an Oops when disconnecting two Gigaset
devices in a row, because the same capi_driver structure gets
unregistered twice. Fix by making driver registration/unregistration
a separate operation (empty in the ISDN4Linux case) called when the
main module is loaded/unloaded.

Impact: bugfix
Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Acked-by: Karsten Keil <keil@b1-systems.de>
CC: stable@kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoobsolete config in kernel source: HSO_AUTOPM
Christoph Egger [Thu, 11 Mar 2010 12:07:54 +0000]
obsolete config in kernel source: HSO_AUTOPM

CONFIG_HSO_AUTOPM is set by KConfig / set in the Kernel source, makefiles
and won't be ever set this way, therefor simply removing the protected
code.

Signed-off-by: Christoph Egger <siccegge@stud.informatik.uni-erlangen.de>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoobsolete config in kernel source: USE_INTERNAL_TIMER
Christoph Egger [Thu, 11 Mar 2010 12:07:52 +0000]
obsolete config in kernel source: USE_INTERNAL_TIMER

CONFIG_USE_INTERNAL_TIMER seems to be the remainings of some experiment.
It is explicitely #undef-ed as not working, only referenced from one
source file and rather aged.

Hereby cleaning it from the kernel tree.

Signed-off-by: Christoph Egger <siccegge@stud.informatik.uni-erlangen.de>
Acked-by: Roel Kluin <roel.kluin@gmail.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoatm: use for_each_set_bit()
Akinobu Mita [Thu, 11 Mar 2010 12:07:50 +0000]
atm: use for_each_set_bit()

Replace open-coded loop with for_each_set_bit().

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agophonet: use for_each_set_bit()
Akinobu Mita [Thu, 11 Mar 2010 12:07:49 +0000]
phonet: use for_each_set_bit()

Replace open-coded loop with for_each_set_bit().

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agotimestamping: fix example build
Randy Dunlap [Fri, 12 Mar 2010 19:23:27 +0000]
timestamping: fix example build

Fix Makefiles so that Documentation/networking/timestamping/timestamping.c
will build when using the CONFIG_BUILD_DOCSRC kconfig option.
(timestamping.c does not build currently with its simple Makefile.)

Also fix printf format warnings.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Patrick Ohly <patrick.ohly@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agodrivers/net: drop redundant memset
Julia Lawall [Tue, 9 Mar 2010 11:14:11 +0000]
drivers/net: drop redundant memset

The region set by the call to memset is immediately overwritten by the
subsequent call to memcpy.

The semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)

// <smpl>
@@
expression e1,e2,e3,e4;
@@

- memset(e1,e2,e3);
  memcpy(e1,e4,e3);
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoi4l: change magic numbers in Eicon DIVA ISDN driver to symbolic names
Ian Munsie [Thu, 11 Mar 2010 12:07:25 +0000]
i4l: change magic numbers in Eicon DIVA ISDN driver to symbolic names

Replace references to the '20' magic number found throughout the Eicon
ISDN driver for the length of the station_id field in the T30_INFO struct
with the T30_MAX_STATION_ID_LENGTH symbolic constant.

Signed-off-by: Ian Munsie <imunsie@au.ibm.com>
Cc: Armin Schindler <mac@melware.de>
Cc: Karsten Keil <isdn@linux-pingi.de>
Cc: Stoyan Gaydarov <sgayda2@uiuc.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoi4l: silence compiler warnings for array access in Eicon DIVA ISDN driver
Ian Munsie [Thu, 11 Mar 2010 12:07:24 +0000]
i4l: silence compiler warnings for array access in Eicon DIVA ISDN driver

When compiling this driver, the compiler throws the following warnings:

drivers/isdn/hardware/eicon/message.c:8426: warning: array subscript is above array bounds
drivers/isdn/hardware/eicon/message.c:8427: warning: array subscript is above array bounds
drivers/isdn/hardware/eicon/message.c:8434: warning: array subscript is above array bounds
drivers/isdn/hardware/eicon/message.c:8435: warning: array subscript is above array bounds
drivers/isdn/hardware/eicon/message.c:8436: warning: array subscript is above array bounds
drivers/isdn/hardware/eicon/message.c:8447: warning: array subscript is above array bounds

This arises from the particular semantics the driver is using to write to
the nlc array (static byte[256]).  The array has a length in byte 0
followed by a T30_INFO struct starting at byte 1.

The T30_INFO struct has a number of variable length strings after the
station_id entry, which cannot be explicitly defined in the struct and the
driver accesses them with an array index to station_id beyond the length
of station_id.

This patch merely changes the semantics that the driver uses to access the
entries after the station_id entry to use the original 256 byte nlc array
taking the offset and length of the station_id entry to calculate where to
write in the array, thereby silencing the warning.

Signed-off-by: Ian Munsie <imunsie@au.ibm.com>
Cc: Armin Schindler <mac@melware.de>
Cc: Karsten Keil <isdn@linux-pingi.de>
Cc: Stoyan Gaydarov <sgayda2@uiuc.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agoisdn: misplaced parenthesis in pof_handle_data()
Roel Kluin [Thu, 11 Mar 2010 12:07:22 +0000]
isdn: misplaced parenthesis in pof_handle_data()

The parenthesis was misplaced.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Cc: Karsten Keil <isdn@linux-pingi.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 years agobnx2x: change smp_mb() comment to conform the true
Stanislaw Gruszka [Tue, 9 Mar 2010 06:55:02 +0000]
bnx2x: change smp_mb() comment to conform the true

Access to fp->tx_bp_prod is protected by __netif_tx_lock,
smp_mb() is not needed for that.

Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>