Calling mod_timer from the rx/tx hotpath is somewhat expensive, and the
timeout doesn't need to be so precise.

Switch to a different strategy: Schedule the timer initially, store jiffies
of all last rx/tx activity which would previously modify the timer, and
let the timer re-arm itself after checking the last rx/tx timestamp.
Make the session timers deferrable to avoid causing extra wakeups on systems
running on battery.
This visibly reduces CPU load under high network load on small embedded
systems.
Signed-off-by: NFelix Fietkau <nbd@openwrt.org>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

  Currently code allows three (HT_AGG_MAX_RETRIES) unanswered addba
requests. When this limit is reached aggregation is turned off for
given TID permanently. This doesn't seem right: three requests is
not that much, some 'blackout' can happen, but effect of it affects
whole connection indefinitely.
  This patch increases number of retries to 15. Also, when there have
been 3 or more retries it splits further retries apart by 15 seconds
instead of sending them in very short period of time.
Signed-off-by: NNikolay Martynov <mar.kolya@gmail.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

Currently BAR, ADDBA and DELBA frames are always sent using AC_VO. If
the TID for which a BA session is established is assigned to a different
queue BAR, ADDBA and DELBA frames can "overtake" frames of the according
BA session.

Hence, always put BA session related frames into the same queue as the
BA sessions data frames.
Signed-off-by: NHelmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

When a station leaves suddenly while ampdu traffic to that station is still
running, there is a possibility that the ampdu pending queues are not freed due
to a race condition leading to memory leaks. In '__sta_info_destroy' when we
attempt to destroy the ampdu sessions in 'ieee80211_sta_tear_down_BA_sessions',
the driver calls 'ieee80211_stop_tx_ba_cb_irqsafe' to delete the ampdu
structures (tid_tx) and splice the pending queues and this job gets queued in
sdata workqueue. However, the sta entry can get destroyed before the above work
gets scheduled and hence the race.

Purging the queues and freeing the tid_tx to avoid the leak. The better solution
would be to fix the race, but that can be taken up in a separate patch.
Signed-off-by: NNishant Sarmukadam <nishants@marvell.com>
Signed-off-by: NYogesh Ashok Powar <yogeshp@marvell.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

Emmanuel noticed that when mac80211 stops the queues
for aggregation that can leave a packet pending. This
packet will be given to the driver after the AMPDU
callback, but as a non-aggregated packet which messes
up the sequence number etc.

I also noticed by looking at the code that if packets
are being processed while we clear the WANT_START bit,
they might see it cleared already and queue up on
tid_tx->pending. If the driver then rejects the new
aggregation session we leak the packet.

Fix both of these issues by changing this code to not
stop the queues at all. Instead, let packets queue up
on the tid_tx->pending queue instead of letting them
get to the driver, and add code to recover properly
in case the driver rejects the session.

(The patch looks large because it has to move two
functions to before their new use.)

Cc: stable@vger.kernel.org
Reported-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

The HT mode is set by iw (previous patchsets).
The interface is set into the specified HT mode.
HT mode and capabilities are announced in beacons.

If we add a station that uses HT also, the fastest matching HT mode will
be used for transmission. That means if we are using HT40+ and we add a station
running on HT40-, we would transfer at HT20.

If we join an IBSS with HT40, but the secondary channel is not
available, we will fall back into HT20 as well.

Allow frame aggregation to start in IBSS mode.
Signed-off-by: NAlexander Simon <an.alexsimon@googlemail.com>
[siwu@hrz.tu-chemnitz.de: Updates]
* remove implicit channel_type enum assumptions
* use rate_control_rate_init() if channel type changed
* remove channel flags check
* activate HT IBSS feature support
* slightly reword commit message
* rebase on wireless-testing
Signed-off-by: NSimon Wunderlich <siwu@hrz.tu-chemnitz.de>
Signed-off-by: NMathias Kretschmer <mathias.kretschmer@fokus.fraunhofer.de>
Reviewed-by: NJohannes Berg <johannes@sipsolutions.net>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

* Follow 802.11n-2009 9.13.3.1 for protection mode and ADDBA
* Send ADDBA only to HT STAs - implement 11.5.1.1 partially
Signed-off-by: NSimon Wunderlich <siwu@hrz.tu-chemnitz.de>
Signed-off-by: NMathias Kretschmer <mathias.kretschmer@fokus.fraunhofer.de>
Reviewed-by: NJohannes Berg <johannes@sipsolutions.net>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

This fixes frequent WARN_ONs when using AP VLAN + aggregation, as these vifs
are virtual and not registered with drivers.
Use sta_info_get_bss instead of sta_info_get in aggregation callbacks, so
that these callbacks can find the station entry when called with the AP vif.
Signed-off-by: NFelix Fietkau <nbd@openwrt.org>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

  Currently tx aggregation is not being timed out even if timeout is
specified when aggregation is opened. Tx tid stays active until delba
arrives from recipient (i.e. recipient times out tid when it is
inactive).
  The problem with this approach is that delba can get lost in the air
and tx tid will stay perpetually opened on the originator while closed
on recipient thus all data sent via this tid will be lost.
  This patch implements tx tid timeouting in way very similar to rx tid
timeouting.
Signed-off-by: NNikolay Martynov <mar.kolya@gmail.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

When a packet is supposed to sent be as an a-MPDU, mac80211 sets
IEEE80211_TX_CTL_AMPDU to let the driver know. On the other
hand, mac80211 configures the driver for aggregration with the
ampdu_action callback.
There is race between these two mechanisms since the following
scenario can occur when the BA agreement is torn down:

Tx softIRQ	 			drv configuration
==========				=================

check OPERATIONAL bit
Set the TX_CTL_AMPDU bit in the packet

					clear OPERATIONAL bit
					stop Tx AGG
Pass Tx packet to the driver.

In that case the driver would get a packet with TX_CTL_AMPDU set
although it has already been notified that the BA session has been
torn down.

To fix this, we need to synchronize all the Qdisc activity after we
cleared the OPERATIONAL bit. After that step, all the following
packets will be buffered until the driver reports it is ready to get
new packets for this RA / TID. This buffering allows not to run into
another race that would send packets with TX_CTL_AMPDU unset while
the driver hasn't been requested to tear down the BA session yet.

This race occurs in practice and iwlwifi complains with a WARN_ON
when it happens.

Cc: stable@kernel.org
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>
Reviewed-by: NJohannes Berg <johannes@sipsolutions.net>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

If addBA responses comes in just after addba_resp_timer has
expired mac80211 will still accept it and try to open the
aggregation session. This causes drivers to be confused and
in some cases even crash.

This patch fixes the race condition and makes sure that if
addba_resp_timer has expired addBA response is not longer
accepted and we do not try to open half-closed session.

Cc: stable@vger.kernel.org
Signed-off-by: NNikolay Martynov <mar.kolya@gmail.com>
[some adjustments]
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

Nikolay noticed (by code review) that mac80211 can
attempt to stop an aggregation session while it is
already being stopped. So to fix it, check whether
stop is already being done and bail out if so.

Also move setting the STOPPING state into the lock
so things are properly atomic.

Cc: stable@vger.kernel.org
Reported-by: NNikolay Martynov <mar.kolya@gmail.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

Signed-off-by: NThomas Pedersen <thomas@cozybit.com>
Signed-off-by: NAshok Nagarajan <anagar6@uic.edu>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

These files are non modular, but need to export symbols using
the macros now living in export.h -- call out the include so
that things won't break when we remove the implicit presence
of module.h from everywhere.
Signed-off-by: NPaul Gortmaker <paul.gortmaker@windriver.com>

The flaglock in struct sta_info has long been
something that I wanted to get rid of, this
finally does the conversion to atomic bitops.

The conversion itself is straight-forward in
most places, a few things needed to change a
bit since we can no longer use multiple bits
at the same time.

On x86-64, this is a fairly significant code
size reduction:
   text	   data	    bss	    dec	    hex
 427861	  23648	   1008	 452517	  6e7a5	before
 425383	  23648	    976	 450007	  6ddd7	after
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

To properly maintain the peer's block ack window, the driver needs to be
able to control the new starting sequence number that is sent along with
the BlockAckReq frame.
Signed-off-by: NFelix Fietkau <nbd@openwrt.org>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

When this flag is set, Tx A-MPDU sessions will not be started by
mac80211. This flag is required for devices that support Tx A-MPDU setup
in hardware.
Signed-off-by: NArik Nemtsov <arik@wizery.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

Removing unnecessary messages saves code and text.

Site specific OOM messages are duplications of a generic MM
out of memory message and aren't really useful, so just
delete them.
Signed-off-by: NJoe Perches <joe@perches.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

While at it also fix the indention of the other IEEE80211_BAR_CTRL_ defines.
Signed-off-by: NHelmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

If we receive an ADDBA response with status code 0 and a buf_size of 0
we should stop the TX BA session as otherwise we'll end up queuing
frames in ieee80211_tx_prep_agg forever instead of sending them out as
non AMPDUs.

This fixes a problem with AVM Fritz Stick N wireless devices where
frames to this device are not transmitted anymore by mac80211.
Signed-off-by: NHelmut Schaa <helmut.schaa@googlemail.com>
Acked-by: NJohannes Berg <johannes@sipsolutions.net>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

This adds sparse RCU annotations to most of
mac80211, only the mesh code remains to be
done.

Due the the previous patches, the annotations
are pretty simple. The only thing that this
actually changes is removing the RCU usage of
key->sta in debugfs since this pointer isn't
actually an RCU-managed pointer (it only has
a single assignment done before the key even
goes live). As that is otherwise harmless, I
decided to make it part of this patch.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

During my quest to make mac80211 not have any RCU
warnings from sparse, I came across the a-MPDU code
again and it wasn't quite clear why it isn't racy.
So instead of assigning the tid_tx array with just
the spinlock held in ieee80211_start_tx_ba_session
use a separate temporary array protected only by
the spinlock and protect all assignments to the
"live" array by both the spinlock and the mutex so
that other code is easily verified to be correct.

Due to pointer assignment atomicity I don't think
this is a real issue, but I'm not sure, especially
on Alpha the current code might be problematic.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

The rcu callback kfree_tid_tx() just calls a kfree(),
so we use kfree_rcu() instead of the call_rcu(kfree_tid_tx).
Signed-off-by: NLai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: NPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: "John W. Linville" <linville@tuxdriver.com>
Cc: Johannes Berg <johannes@sipsolutions.net>
Reviewed-by: NJosh Triplett <josh@joshtriplett.org>
Acked-by: N"David S. Miller" <davem@davemloft.net>

Currently, mac80211 always advertises that it may send
up to 64 subframes in an aggregate. This is fine, since
it's the max, but might as well be set to zero instead
since it doesn't have any information.

However, drivers might have that information, so allow
them to set a variable giving it, which will then be
used. The default of zero will be fine since to the
peer that means we don't know and it will just use its
own limit for the buffer size.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

The aggregation code currently doesn't implement the
buffer size negotiation. It will always request a max
buffer size (which is fine, if a little pointless, as
the mac80211 code doesn't know and might just use 0
instead), but if the peer requests a smaller size it
isn't possible to honour this request.

In order to fix this, look at the buffer size in the
addBA response frame, keep track of it and pass it to
the driver in the ampdu_action callback when called
with the IEEE80211_AMPDU_TX_OPERATIONAL action. That
way the driver can limit the number of subframes in
aggregates appropriately.

Note that this doesn't fix any drivers apart from the
addition of the new argument -- they all need to be
updated separately to use this variable!
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

Allow drivers or rate control algorithms to specify BlockAck session
timeout when initiating an ADDBA transaction. This is useful in cases
where maintaining persistent BA sessions does not incur any overhead.

The current timeout value of 5000 TUs is retained for all non ath9k/ath9k_htc
drivers.
Signed-off-by: NSujith Manoharan <Sujith.Manoharan@atheros.com>
Reviewed-by: NJohannes Berg <johannes@sipsolutions.net>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

When roaming while we have active BA session,
we can end up transmitting delBA frames to
the old AP while we're already on the new AP's
channel, which can cause warnings.

Simply avoid sending those frames, but still
tear down the internal session state, since
they are not really necessary anyway as we
will implicitly disassociate when sending the
association to the new AP.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Acked-by: NLuis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

We never delete the addBA response timer, which
is typically fine, but if the station it belongs
to is deleted very quickly after starting the BA
session, before the peer had a chance to reply,
the timer may fire after the station struct has
been freed already. Therefore, we need to delete
the timer in a suitable spot -- best when the
session is being stopped (which will happen even
then) in which case the delete will be a no-op
most of the time.

I've reproduced the scenario and tested the fix.

This fixes the crash reported at
http://mid.gmane.org/4CAB6F96.6090701@candelatech.com

Cc: stable@kernel.org
Reported-by: NBen Greear <greearb@candelatech.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

Even before the recent changes, the documentation
for TX aggregation was somewhat out of date. Update
it and also add documentation for the RX side.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

To prepare for allowing drivers to sleep in
ampdu_action, change the locking in the TX
aggregation code to use the mutex the RX part
already uses. The spinlock is still necessary
around some code to avoid races with TX, but
now we can also synchronize_net() to avoid
getting an inconsistent sequence number.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

Since we want the code to be able to sleep
in the future, it must not be called from
the timer directly. To achieve that, simply
call the function drivers would call, and
also use RCU in the timer to get the struct
so we don't need to rely on the spinlock in
the future.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

Move the block-ack session works into common
code, since it will be needed for RX agg too
in the next patches.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

When the driver or rate control requests starting
or stopping an aggregation session, that currently
causes a direct callback into the driver, which
could potentially cause locking problems. Also,
the functions need to be callable from contexts
that cannot sleep, and thus will interfere with
making the ampdu_action callback sleeping.

To address these issues, add a new work item for
each station that will process any start or stop
requests out of line.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

mac80211 currently maintains the ampdu_lock to
avoid starting a queue due to one aggregation
session while another aggregation session needs
the queue stopped.

We can do better, however, and instead refcount
the queue stops for this particular purpose,
thus removing the need for the lock. This will
help making ampdu_action able to sleep.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

The non-irqsafe aggregation start/stop done
callbacks are currently only used by ath9k_htc,
and can cause callbacks into the driver again.
This might lead to locking issues, which will
only get worse as we modify locking. To avoid
trouble, remove the non-irqsafe versions and
change ath9k_htc to use those instead.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

Currently we allocate some memory for each TX
aggregation session and additionally keep a
state bitmap indicating the state it is in.
By using RCU to protect the pointer, moving
the state into the structure and some locking
trickery we can avoid locking when the TX agg
session is fully operational.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

This moves the aggregation callback processing
to the per-sdata skb queue and a work function
rather than the tasklet.

Unfortunately, this means that it extends the
pkt_type hack to that skb queue. However, it
will enable making ampdu_action API changes
gradually, my current plan is to get rid of
this again by forcing drivers to only return
from ampdu_action() when everything is done,
thus removing the callbacks completely.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

A number of places use RCU locking for accessing
the station list, even though they do not need
to. Use mutex locking instead to prepare for the
locking changes I want to make. The mlme code is
also using a WLAN_STA_DISASSOC flag that has the
same meaning as WLAN_STA_BLOCK_BA, so use that.

While doing so, combine places where we loop
over stations twice, and optimise away some of
the loops by checking if the hardware supports
aggregation at all first.

Also fix a more theoretical race condition: right
now we could resume, set up an aggregation session,
and right after tear it down again due to the code
that is needed for hardware reconfiguration here.
Also mark add a comment to that code marking it as
a workaround.

Finally, remove a pointless aggregation disabling
loop when an interface is stopped, directly after
that we remove all stations from it which will also
disable all aggregation sessions that may still be
active, and does so in a race-free way unlike the
current loop that doesn't block new sessions.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

There's no sense in letting anything but internal
mac80211 functions set the initiator to anything
but WLAN_BACK_INITIATOR, since WLAN_BACK_RECIPIENT
is only valid when we have received a frame from
the peer, which we react to directly in mac80211.

The debugfs code I recently added got this wrong
as well.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

The dialog token allocator has apparently been broken
since b83f4e15 ("mac80211: fix deadlock in sta->lock")
because it got moved out under the spinlock. Fix it.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>