mac80211: clean up rate control API

Long awaited, hard work. This patch totally cleans up the rate control
API to remove the requirement to include internal headers outside of
net/mac80211/.

There's one internal use in the PID algorithm left for mesh networking,
we'll have to figure out a way to clean that one up and decide how to
do the peer link evaluation, possibly independent of the rate control
algorithm or via new API.

Additionally, ath9k is left using the cross-inclusion hack for now, we
will add new API where necessary to make this work properly, but right
now I'm not expert enough to do it. It's still off better than before.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

mac80211: fix mesh action frame handling

When I split off the action frame handling I made the code drop
all action frames we don't want to handle. This is wrong since
some action frames are actually handled via rx_h_mgmt through
being queued to the sta/mesh implementations.

Thanks to Li YanBo for noticing the problem.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Cc: Li YanBo <dreamfly281@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

mac80211: mesh portal functionality support

Currently the mesh code doesn't support bridging mesh point interfaces
with wired ethernet or AP to construct an MPP or MAP. This patch adds
code to support the "6 address frame format packet" functionality to
mesh point interfaces. Now the mesh network can be used as backhaul
for end to end communication.

Signed-off-by: Li YanBo <dreamfly281@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath9k: Fix TSF Adjust usage

TSF adjust is needed only for AP mode when staggered beacons
are used. Since we support only a single interface in IBSS mode,
disable setting the TSF adjust register.

Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath9k: Fix nexttbtt calculation

nexttbtt has to be obtained from the timestamp of the beacon
obtained from mac80211. Fix this.
And <asm/unaligned.h> is not needed anymore.

Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath9k: Whitespace cleanup

Also, remove comments that are not relevant anymore.

Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: remove HT flags from RXON when not in HT anymore

This patch removes the HT flags from RXON when moving from HT to legacy.
This avoids keeping those flags set and possibly miss configuring firmware.

If we are configured in HT, fat channel: channel 1 above, and move later
to legacy channel 11, we need to clear the FAT channel control flags in
RXON. If we don't, the firmware will understand this as channel 11 above
which is not possible due to regulatory constraints, leading to firmware
crash.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Reviewed-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

libertas: Reduce the WPA key installation time (fixups)

This patch addresses comments from Dan Williams about the patch
committed as "libertas: Reduce the WPA key installation time."

Signed-off-by: Javier Cardona <javier@cozybit.com>
Acked-by: Dan Williams <dcbw@redhat.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath5k: Fix bad udelay calls on AR5210 code

 * Fix bad udelay calls (using > 2000us) in AR5210 code and clean up
 some bits on nic_reset (AR5210 support is still in bad shape)

 Changes-licensed-under: ISC
Signed-off-by: Nick Kossifidis <mickflemm@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

cfg80211: reject invalid configuration items

Reject configuring mesh-id for non-mesh, monitor flags for non-monitor.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

mac80211: allow interface settings changes only when down

We currently allow monitor flags changes and mesh ID changes when
the interface is up, which can lead to trouble. Change it to only
allow when down.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

cfg80211: allow set_interface without type

Which then causes no type change.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

cfg80211: show interface type

This patch makes cfg80211 show the interface in the nl80211
information about a specific interface. API users are required
to keep the type updated (everything else is fairly complicated)
but you will get a warning if you fail to keep it updated.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

mac80211: make master iface not wireless

There's no need to register the master netdev with cfg80211,
in fact, this is quite dangerous and lead to having to add
checks for the master interface all over the config handlers.
This patch removes the "ieee80211_ptr" from the master iface
in favour of having a small netdev_priv() associated with
the master interface that stores the ieee80211_local pointer.
Because of this, a lot of code in the configuration handlers
can go away. To make this patch easier to verify I have also
removed a number of wiphy_priv() calls in favour of getting
the sdata first and then the local pointer from that.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: make PS use named constants

This patch adds named constants for configuring MIMO power save
chain settings.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: don't delay scan in IBSS mode

Scan need to be delayed only after association to allow EAPOL
exchange. We don't need the delay for IBSS mode.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

b43: Issue warning when RFKILL_INPUT is not enabled

If the system is misconfigured with CONFIG_RFKILL set but CONFIG_RFKILL_INPUT
not set, the built-in radio LEDs will not work. In the current code, no warning
is issued.

Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

cfg80211: fix code ordering in header file

Luis added the regulatory hint stuff to this file without
observing that __ieee80211_get_channel and ieee80211_get_channel
really belong together.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

cfg80211: clean up static regdomain mess

The statically defined regdomains are used in a very convoluted
way, use them instead to prime the information we have and then
continue operating normally.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

cfg80211: fix regulatory code const

A few pointers and structures in the regulatory code are const,
but because it wasn't done properly a whole bunch of bogus
casts were needed to compile without warning. Mark everything
const properly to avoid that kind of junk code.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

cfg80211: clean up regulatory mess

The recent code from Luis is an #ifdef hell and contains lots of
code that's stuffed into the wrong file making a whole bunch of
things needlessly non-static, and besides, what is it doing in
core.c??

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

cfg80211: fix static regdomains

When Luis added the static regdomains back he used +/-20
of the centre frequencies to account for 40MHz bandwidth
neglecting the fact that 40MHz bandwidth cannot be used
on the channels close to the allowed band edges.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

wireless: Read scan flags correctly on x86-64

The SIOCSIWSCAN handler is passed data in an iw_point structure. Some
drivers erronously use an iw_param instead.

On 32 bit architectures the difference isn't noticed as the flags
parameter tends to be the only one used by scan handlers and is at the
same offset.

On 64 bit architectures the pointer in the iw_point structure means the
flag parameter is at different offsets in these structures.

Thanks to Jean Tourrilhes for tracking this down for orinoco, and Pavel
Roskin for confirming the fix and identifying other suspect handlers.

Signed-off-by: David Kilroy <kilroyd@googlemail.com>
Acked-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

aoe: Fix OOPS after SKB queue changes.

Reported by Thomas Graf.

If we don't unlink the SKB from the queue when we send it
out in aoenet_xmit(), dev_hard_start_xmit() will see skb->next
as non-NULL and interpret this to mean the SKB is part of a
GSO segment list.

Add __skb_unlink() call to fix that.

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

can: Add documentation for virtual CAN driver usage

This patch adds a usage documentation for the virtual CAN driver (vcan).

Signed-off-by: Oliver Hartkopp <oliver@hartkopp.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: Fix queue traversal in tcp_use_frto().

We must check tcp_skb_is_last() before doing a tcp_write_queue_next().

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

tcp: Fix order of tests in tcp_retransmit_skb()

tcp_write_queue_next() must only be made if we know that
tcp_skb_is_last() evaluates to false.

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

ppp_generic: Use skb_peek() in ppp_receive_mp_frame().

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

neigh: Remove by-hand SKB queue handling.

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

pkt_sched: Remove the tx queue state check in qdisc_run()

The current check wrongly uses the state of one (currently the first)
tx queue for all tx queues in case of non-default qdiscs. This check
mainly prevented requeuing loop with __netif_schedule(), but now it's
controlled inside __qdisc_run(), while dequeuing. The wrongness of
this check was first noticed by Herbert Xu.

Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: Use SKB queue handling interfaces instead of by-hand versions.

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

net: Add skb_queue_walk_from() and skb_queue_walk_from_safe().

These will be used by TCP write queue handling and elsewhere.

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

net: Add skb_queue_next().

A lot of code wants to iterate over an SKB queue at the top level using
it's own control structure and iterator scheme.

Provide skb_queue_next(), which is only valid to invoke if
skb_queue_is_last() returns false.

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

tcp: Use skb_queue_is_last() instead of by-hand version.

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

net: Add skb_queue_is_last().

Several bits of code want to know "is this the last SKB in
a queue", and all of them implement this by hand.

Provide an common interface to make this check.

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

usbatm: Use skb_queue_walk_safe() instead of by-hand implementation.

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

zd1211: Use skb_queue_walk() instead of by-hand implementation.

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

de4x5: Use skb_queue_head instead of by-hand implementation.

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

pkt_sched: Check the state of tx_queue in dequeue_skb()

Check in dequeue_skb() the state of tx_queue for requeued skb to save
on locking and re-requeuing, and possibly remove the current check in
qdisc_run(). Based on the idea of Alexander Duyck.

Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

pkt_sched: Always use q->requeue in dev_requeue_skb().

There is no reason to call into the complicated qdiscs
just to remember the last SKB where we found the device
blocked.

The SKB is outside of the qdiscs realm at this point.

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

pkt_sched: Make qdisc->gso_skb a list.

The idea is that we can use this to get rid of
->requeue().

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

sctp: Use skb_queue_walk_safe() and skb_queue_split_tail_init().

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

net: Fix bus in SKB queue splicing interfaces.

Handle the case of head being non-empty, by adding list->qlen
to head->qlen instead of using direct assignment.

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

net: network device name ifalias support

This patch add support for keeping an additional character alias
associated with an network interface. This is useful for maintaining
the SNMP ifAlias value which is a user defined value. Routers use this
to hold information like which circuit or line it is connected to. It
is just an arbitrary text label on the network device.

There are two exposed interfaces with this patch, the value can be
read/written either via netlink or sysfs.

This could be maintained just by the snmp daemon, but it is more
generally useful for other management tools, and the kernel is good
place to act as an agreed upon interface to store it.

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

usbnet: Use skb_queue_walk_safe() instead of by-hand implementation.

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

Phonet: kernel documentation

Signed-off-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: emit errors when a packet cannot be delivered locally

When there is no listener socket for a received packet, send an error
back to the sender.

Signed-off-by: Remi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: proc interface for port range

Phonet endpoints are bound to individual ports.
This provides a /proc/sys/net/phonet (or sysctl) interface for
selecting the range of automatically allocated ports (much like the
ip_local_port_range with IPv4).

Signed-off-by: Remi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: provide MAC header operations

Signed-off-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: Phonet datagram transport protocol

This provides the basic SOCK_DGRAM transport protocol for Phonet.

Signed-off-by: Remi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: common socket glue

This provides the socket API for the Phonet protocols family.

Signed-off-by: Remi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: Netlink interface

This provides support for configuring Phonet addresses, notifying
Phonet configuration changes, and dumping the configuration.

Signed-off-by: Remi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: network device and address handling

This provides support for adding Phonet addresses to and removing
Phonet addresses from network devices.

Signed-off-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: add CONFIG_PHONET

Signed-off-by: Remi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: PF_PHONET protocol family support

This is the basis for the Phonet protocol families, and introduces
the ETH_P_PHONET packet type and the PF_PHONET socket family.

Signed-off-by: Remi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: global definitions

Signed-off-by: Remi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipsec: Fix xfrm_state_walk race

As discovered by Timo Teräs, the currently xfrm_state_walk scheme
is racy because if a second dump finishes before the first, we
may free xfrm states that the first dump would walk over later.

This patch fixes this by storing the dumps in a list in order
to calculate the correct completion counter which cures this
problem.

I've expanded netlink_cb in order to accomodate the extra state
related to this.  It shouldn't be a big deal since netlink_cb
is kmalloced for each dump and we're just increasing it by 4 or
8 bytes.

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

cassini: use request_firmware

Firmware blob looks like this...
        __le16 load_address
        unsigned char data[]

[akpm@linux-foundation.org: include vmalloc.h]
Signed-off-by: Jaswinder Singh <jaswinder@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/atm/lec.c: drop code after return

The break after the return serves no purpose.

Signed-off-by: Julia Lawall <julia@diku.dk>
Reviewed-by: Richard Genoud <richard.genoud@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/wan: Use DIV_ROUND_UP

The kernel.h macro DIV_ROUND_UP performs the computation (((n) + (d) - 1) /
(d)) but is perhaps more readable.

An extract of the semantic patch that makes this change is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@haskernel@
@@

#include <linux/kernel.h>

@depends on haskernel@
expression n,d;
@@

(
- (n + d - 1) / d
+ DIV_ROUND_UP(n,d)
|
- (n + (d - 1)) / d
+ DIV_ROUND_UP(n,d)
)

@depends on haskernel@
expression n,d;
@@

- DIV_ROUND_UP((n),d)
+ DIV_ROUND_UP(n,d)

@depends on haskernel@
expression n,d;
@@

- DIV_ROUND_UP(n,(d))
+ DIV_ROUND_UP(n,d)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/atm: Use DIV_ROUND_UP

The kernel.h macro DIV_ROUND_UP performs the computation (((n) + (d) - 1) /
(d)) but is perhaps more readable.

In the case of the file drivers/atm/eni.c, I am a little bit suspicious of
the -1 at the end of the affected expression.  Please check that that is
what is wanted.

An extract of the semantic patch that makes this change is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@haskernel@
@@

#include <linux/kernel.h>

@depends on haskernel@
expression n,d;
@@

(
- (n + d - 1) / d
+ DIV_ROUND_UP(n,d)
|
- (n + (d - 1)) / d
+ DIV_ROUND_UP(n,d)
)

@depends on haskernel@
expression n,d;
@@

- DIV_ROUND_UP((n),d)
+ DIV_ROUND_UP(n,d)

@depends on haskernel@
expression n,d;
@@

- DIV_ROUND_UP(n,(d))
+ DIV_ROUND_UP(n,d)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

hci_usb: replace mb with smp_mb

smp_mb is enough for ordering memory operations among processors,and mb is
more expensive than smp_mb for UP machine, so replace it with smp_mb().

Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Acked-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: em_cmp.c use unaligned access helpers

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

irda: follow lspci device/vendor style

Use "[%04x:%04x]" for PCI vendor/device IDs to follow the format used by
lspci(8).

Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

mISDN: misc timerdev fixes

- Remove noop VFS stubs. The VFS does that on a NULL pointer anyways.
- Fix timer handler prototype to be correct
- Comment ugly SMP race I didn't fix.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

mISDN: annotate iomem pointer and add statics

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

mISDN: endian annotations for struct zt

Found two possible bugs where the z1 value was used directly without
byteswapping.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/isdn/capi/kcapi.c: Adjust error handling code involving capi_ctr_put

After calling capi_ctr_get, error handling code should call capi_ctr_put.

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

// <smpl>
@r@
expression x,E;
statement S;
position p1,p2,p3;
@@

(
if ((x = capi_ctr_get@p1(...)) == NULL || ...) S
|
x = capi_ctr_get@p1(...)
... when != x
if (x == NULL || ...) S
)
<...
if@p3 (...) { ... when != capi_ctr_put(x)
                  when != if (x) { ... capi_ctr_put(x); ...}
    return@p2 ...;
}
...>
(
return x;
|
return 0;
|
x = E
|
E = x
|
capi_ctr_put(x)
)

@exists@
position r.p1,r.p2,r.p3;
expression x;
int ret != 0;
statement S;
@@

* x = capi_ctr_get@p1(...)
  <...
* if@p3 (...)
  S
  ...>
* return@p2 \(NULL\|ret\);
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

netfilter: xt_time gives a wrong monthday in a leap year

The function localtime_3 in xt_time.c gives a wrong monthday in a leap
year after 28th 2.  calculating monthday should use the array
days_since_leapyear[] not days_since_year[] in a leap year.

Signed-off-by: Kaihui Luo <kaih.luo@gmail.com>
Acked-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb3: Use SKB list interfaces instead of home-grown implementation.

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

isdn: isdn_ppp: Use SKB list facilities instead of home-grown implementation.

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

bluetooth: hci_bcsp: Use SKB list interfaces instead of home-grown stuff.

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

aoe: Use SKB interfaces for list management instead of home-grown stuff.

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

net: Add new interfaces for SKB list light-weight init and splicing.

This will be used by subsequent changesets.

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

atm: idt77252: Use generic SKB queue management instead of home-grown scheme.

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

net: Remove __skb_insert() calls outside of skbuff internals.

This minor cleanup simplifies later changes which will convert
struct sk_buff and friends over to using struct list_head.

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

tcp: advertise MSS requested by user

I'm trying to use the TCP_MAXSEG option to setsockopt() to set the MSS
for both sides of a bidirectional connection.

man tcp says: "If this option is set before connection establishment, it
also changes the MSS value announced to the other end in the initial
packet."

However, the kernel only uses the MTU/route cache to set the advertised
MSS. That means if I set the MSS to, say, 500 before calling connect(),
I will send at most 500-byte packets, but I will still receive 1500-byte
packets in reply.

This is a bug, either in the kernel or the documentation.

This patch (applies to latest net-2.6) reduces the advertised value to
that requested by the user as long as setsockopt() is called before
connect() or accept(). This seems like the behavior that one would
expect as well as that which is documented.

I've tried to make sure that things that depend on the advertised MSS
are set correctly.

Signed-off-by: Tom Quetchenbach <virtualphtn@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Use hton[sl]() instead of __constant_hton[sl]() where applicable

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

multiq: requeue should rewind the current_band

Currently dequeueing a packet and requeueing the same packet will cause a
different packet to be pulled on the next dequeue.  This change forces
requeue to rewind the current_band.

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

tcp: back retransmit_high when it over-estimated

If lost skb is sacked, we might have nothing to retransmit
as high as the retransmit_high is pointing to, so place
it lower to avoid unnecessary walking.

This is mainly for the case where high L'ed skbs gets sacked.

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

tcp: don't clear lost_skb_hint when not necessary

Most importantly avoid doing it with cumulative ACK. However,
since we have lost_cnt_hint in the picture as well needing
adjustments, it's not as trivial as dealing with
retransmit_skb_hint (and cannot be done in the all place we
could trivially leave retransmit_skb_hint untouched).

With the previous patch, this should mostly remove O(n^2)
behavior while cumulative ACKs start flowing once rexmit
after a lossy round-trip made it through.

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

tcp: don't clear retransmit_skb_hint when not necessary

Most importantly avoid doing it with cumulative ACK. Not clearing
means that we no longer need n^2 processing in resolution of each
fast recovery.

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

tcp: remove retransmit_skb_hint clearing from failure

This doesn't much sense here afaict, probably never has. Since
fragmenting and collapsing deal the hints by themselves, there
should be very little reason for the rexmit loop to do that.

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

tcp: reorganize retransmit code loops

Both loops are quite similar, so they can be combined
with little effort. As a result, forward_skb_hint becomes
obsolete as well.

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

tcp: remove tp->lost_out guard to make joining diff nicer

The validity of the retransmit_high must then be ensured
if no L'ed skb exits!

This makes a minor change to behavior, we now have to
iterate the head to find out that the loop terminates.

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

tcp: Reorganize skb tagbit checks

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

tcp: remove obsolete validity concern

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

tcp: add tcp_can_forward_retransmit

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

tcp: No need to clear retransmit_skb_hint when SACKing

Because lost counter no longer requires tuning, this is
trivial to remove (the tuning wouldn't have been too
hard either) because no "new" retransmittable skb appeared
below retransmit_skb_hint when SACKing for sure.

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

tcp: Kill precaution that's very likely obsolete

I suspect it might have been related to the changed amount
of lost skbs, which was counted by retransmit_cnt_hint that
got changed.

The place for this clearing was very illogical anyway,
it should have been after the LOST-bit clearing loop to
make any sense.

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

tcp: convert retransmit_cnt_hint to seqno

Main benefit in this is that we can then freely point
the retransmit_skb_hint to anywhere we want to because
there's no longer need to know what would be the count
changes involve, and since this is really used only as a
terminator, unnecessary work is one time walk at most,
and if some retransmissions are necessary after that
point later on, the walk is not full waste of time
anyway.

Since retransmit_high must be kept valid, all lost
markers must ensure that.

Now I also have learned how those "holes" in the
rexmittable skbs can appear, mtu probe does them. So
I removed the misleading comment as well.

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

tcp: add helper for lost bit toggling

This useful because we'd need to verifying soon in many places
which makes things slightly more complex than it used to be.

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

tcp: move tcp_verify_retransmit_hint

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

tcp: Partial hint clearing has again become meaningless

Ie., the difference between partial and all clearing doesn't
exists anymore since the SACK optimizations got dropped by
an sacktag rewrite.

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

Merge branch 'master' of git://git./linux/kernel/git/linville/wireless-next-2.6

qlge: Protect qlge_resume() with CONFIG_PM

Fixes the following build warning:

drivers/net/qlge/qlge_main.c:3897: warning: ‘qlge_resume’ defined but not used

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

qlge: Fix warnings in debugging code.

drivers/net/qlge/qlge_dbg.c: In function ‘ql_dump_qdev’:
drivers/net/qlge/qlge_dbg.c:369: warning: cast to pointer from integer of different size
drivers/net/qlge/qlge_dbg.c:373: warning: cast to pointer from integer of different size
drivers/net/qlge/qlge_dbg.c: In function ‘ql_dump_tx_ring’:
drivers/net/qlge/qlge_dbg.c:457: warning: format ‘%llx’ expects type ‘long long unsigned int’, but argument 2 has type ‘long unsigned int’
drivers/net/qlge/qlge_dbg.c:461: warning: format ‘%llx’ expects type ‘long long unsigned int’, but argument 2 has type ‘long unsigned int’
drivers/net/qlge/qlge_dbg.c: In function ‘ql_dump_rx_ring’:
drivers/net/qlge/qlge_dbg.c:557: warning: format ‘%llx’ expects type ‘long long unsigned int’, but argument 2 has type ‘long unsigned int’
drivers/net/qlge/qlge_dbg.c:565: warning: format ‘%llx’ expects type ‘long long unsigned int’, but argument 2 has type ‘long unsigned int’
drivers/net/qlge/qlge_dbg.c:575: warning: format ‘%llx’ expects type ‘long long unsigned int’, but argument 2 has type ‘long unsigned int’
drivers/net/qlge/qlge_dbg.c:579: warning: format ‘%llx’ expects type ‘long long unsigned int’, but argument 2 has type ‘long unsigned int’
drivers/net/qlge/qlge_dbg.c:598: warning: format ‘%llx’ expects type ‘long long unsigned int’, but argument 2 has type ‘long unsigned int’
drivers/net/qlge/qlge_dbg.c:602: warning: format ‘%llx’ expects type ‘long long unsigned int’, but argument 2 has type ‘long unsigned int’

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

Merge branch 'new-drivers' of /linux/kernel/git/jgarzik/netdev-2.6

net: Fix build with ARCH=um

If UM is going to claim that it supports DMA by setting
HAS_DMA, it should provide a dma_mapping_error() implementation.

Based upon a report by Julius Volz.

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

ISDN sockets: add missing lockdep strings

Signed-off-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>