Instead of using the out-of-line EWMA calculation, use DECLARE_EWMA()
to create static inlines. On x86/64 this results in no change in code
size for me, but reduces the struct receive_queue size by the two
unsigned long values that store the parameters.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This patch adds fix to set carrier state off during ndo_open.
Carrier should be set to ON when device is ready to send data.
In case of station/adhoc interface device is able to transfer
data after successful association/join operation.
For AP this would be after bss_active event.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Avinash Patil <[email protected]>
Signed-off-by: Kalle Valo <[email protected]>
There are better ways to get the kernel information, use the
utsname and omit the version code entirely since it's duplicate.
The version magic is rather useless anyway
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Kalle Valo <[email protected]>
When an interface is deleted, an ongoing hardware scan is canceled and
the driver must abort the scan, at the very least reporting completion
while the interface is removed.
However, if it scheduled the work that might only run after everything
is said and done, which leads to cfg80211 warning that the scan isn't
reported as finished yet; this is no fault of the driver, it already
did, but mac80211 hasn't processed it.
To fix this situation, flush the delayed work when the interface being
removed is the one that was executing the scan.
Cc: [email protected]
Reported-by: Sujith Manoharan <[email protected]>
Tested-by: Sujith Manoharan <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
As reported in https://bugzilla.kernel.org/show_bug.cgi?id=60514,
the station loop never initialises 'sinfo' and therefore adds up
a stack values, leaking stack information (the number of times it
adds values is easily obtained another way.)
Fix this by initialising the sinfo for each station to add.
Cc: [email protected]
Signed-off-by: Johannes Berg <[email protected]>
When splitting the SME implementation from the MLME code,
I introduced a few bugs:
* association failures no longer sent a connect-failure event
* getting disassociated from the AP caused deauth to be sent
but state wasn't cleaned up, leading to warnings
* authentication failures weren't cleaned up properly, causing
new connection attempts to warn and fail
Fix these bugs.
Signed-off-by: Johannes Berg <[email protected]>
Unfortunately, WoL is broken and the system will immediately
resume after suspending, and I can't seem to figure out why.
Remove WoL support until the issue can be found.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
A number of places treated features wrongly, listing not-supported
features instead of supported ones. Also, the get_drvinfo ethtool
callback isn't needed, and alx_get_pauseparam can be simplified.
Reported-by: Ben Hutchings <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
In two places, parts of MAC addresses are used as u32/u16
values. This can cause alignment problems, use put_unaligned
and get_unaligned to fix this.
Reported-by: Ben Hutchings <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
As suggested by Ben Hutchings, use separate fields to track
current link speed and duplex setting.
Reported-by: Ben Hutchings <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The ring sizes should be unsigned, pointed out by Ben Hutchings.
Reported-by: Ben Hutchings <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
100mbit half duplex is ADVERTISED_100baseT_Half, not
ADVERTISED_10baseT_Half.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Even when alx_setup_speed_duplex() is called, we still
need to call alx_cfg_mac_flowcontrol() and set hw->flowctrl
if flow control changed.
This was a bug I accidentally introduced while simplifying
the original driver.
Reported-by: Ben Hutchings <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This is a very simple driver, based on the original vendor
driver that Qualcomm/Atheros published/submitted previously,
but reworked to make the code saner. However, it also lost
a number of features (TSO/GSO, VLAN acceleration and multi-
queue support) in the process, as well as debugging support
features I didn't have any use for. The only thing I left
is checksum offload.
More features can obviously be added, but this seemed like
a good start for having a driver in mainline at all.
Johannes Stezenbach has verified that the driver works on
AR8161, I have a AR8171 myself. The E2200 device ID I found
on github in somebody's repository.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
In the odd case that while updating information from a beacon,
a BSS was found that is part of a hidden group, we drop the
new information. In this case, however, we leak the IE buffer
from the update, and erroneously update the entry's timestamp
so it will never time out. Fix both these issues.
Cc: Larry Finger <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
I've had a new key for almost two years and
revoked the old one, but forgot to update it
here.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Jiri Kosina <[email protected]>
This reverts commit 5a254ffe3f.
The commit failed to take into account that allocated wireless devices
(wiphys) are not added into the device list upon allocation, but only
when they are registered. Therefore, it opened up a race between
allocating and registering a name, so that if two processes allocate and
register concurrently ("alloc, alloc, register, register" rather than
"alloc, register, alloc, register") the code will attempt to use the
same name twice.
Signed-off-by: Johannes Berg <[email protected]>
The new_name argument to device_rename() can be
const as kobject_rename's new_name argument is.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
Ever since
commit e1b3ec1a2a
Author: Stanislaw Gruszka <[email protected]>
Date: Mon Mar 29 12:18:34 2010 +0200
mac80211: explicitly disable/enable QoS
mac80211 is telling drivers, in particular
iwlwifi, whether QoS is enabled or not.
However, this is only relevant for station mode,
since only then will any device send nullfunc
frames and need to know whether they should be
QoS frames or not. In other modes, there are
(currently) no frames the device is supposed to
send.
When you now consider virtual interfaces, it
becomes apparent that the current mechanism is
inadequate since it enables/disables QoS on a
global scale, where for nullfunc frames it has
to be on a per-interface scale.
Due to the above considerations, we can change
the way mac80211 advertises the QoS state to
drivers to only ever advertise it as "off" in
station mode, and make it a per-BSS setting.
Tested-by: Stanislaw Gruszka <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When SPARSE_IRQ is set, irq_to_desc() can
return NULL. While the code here has a
check for NULL, it's not really correct.
Fix it by separating the check for it.
This fixes CPU hot unplug for me.
Reported-by: Alastair Bridgewater <[email protected]>
Cc: [email protected] [2.6.32+]
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Benjamin Herrenschmidt <[email protected]>
Fix a whole bunch of kernel-doc warnings
and errors that crop up when running it on
mac80211 and cfg80211; the latter isn't
normally done so lots of bit-rot happened.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
After ieee80211_rx_h_ctrl() processing we only
want to process management (including action)
frames, so there's no point in letting control
frames continue.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Ever since we use only cfg80211 for configuration,
there is no configuration that could be pending at
this point, cfg80211 will have the configuration
that is pending and apply it afterwards.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
I suspect the compiler will do this optimisation
anyway, but it seems cleaner to move this into
the WEP switch case.
Also make rx_h_decrypt use a local variable for
the frame_control so that we don't need to reload
the hdr variable for this after linearizing.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
There's a single use of this struct member, but
as it is write-only it clearly not necessary.
Thus we can free up some space here, even if we
don't need it right now it seems pointless to
carry around the variable.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
The IBSS code has a bogus mod_timer(..., 0) call,
we shouldn't ever pass a constant value to the
function since any constant value could be in the
future or the past.
However, invoking the timer here is not necessary
at all, since we just finished scanning and just
need to have the IBSS code run again from the
workqueue later, so factor out the work starting
and use that instead.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When sending action frames, we want to verify
that we do that on the correct channel. However,
checking the channel type in addition can get in
the way, since the channel type could change on
the fly during an association, and it's not
useful to have the channel type anyway since it
has no effect on the transmission. Therefore,
make it optional to specify so that if wanted,
it can still be checked, but is not required.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Daniel reported that the paged RX changes had
broken blockack request frame processing due
to using data that wasn't really part of the
skb data.
Fix this using skb_copy_bits() for the needed
data. As a side effect, this adds a check on
processing too short frames, which previously
this code could do.
Reported-by: Daniel Halperin <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Acked-by: Daniel Halperin <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
sparse correctly complains that
__ieee80211_get_channel_mode is not static.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Specifying a valid channel type will get
goto out rather than continuing, due to
missing braces. This affects both remain
on channel and action frame TX commands.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This adds support for offloading the channel switch
operation to devices that support such, typically
by having specific firmware API for it. The reasons
for this could be that the firmware provides better
timing or that regulatory enforcement done by the
device requires special handling of CSAs.
In order to allow drivers to specify the timing to
the device, the new channel_switch callback will
pass through the received frame's mactime, where
available.
Signed-off-by: Wey-Yi Guy <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When we process a frame, we currently just match it
to the work struct by the MAC addresses, and not by
the work type. This means that we can end up doing
the work for an association request item when (for
whatever reason) we receive another frame type, for
example a probe response. Processing the wrong type
of frame will lead to completely invalid data being
processed, and will lead to various problems like
thinking the association was successful even if the
AP never sent an assocation response.
Fix this by making each processing function check
that it is invoked for the right work struct type
only and continue processing otherwise (and drop
frames that we didn't expect).
This bug was uncovered during the debugging for
https://bugzilla.kernel.org/show_bug.cgi?id=15862
but doesn't seem to be the cause for any of the
various problems reported there.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Currently, when one interface switches HT mode,
all others will follow along. This is clearly
undesirable, since the new one might switch to
no-HT while another one is operating in HT.
Address this issue by keeping track of the HT
mode per interface, and allowing only changes
that are compatible, i.e. switching into HT40+
is not possible when another interface is in
HT40-, in that case the second one needs to
fall back to HT20.
Also, to allow drivers to know what's going on,
store the per-interface HT mode (channel type)
in the virtual interface's bss_conf.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Currently (all tested with hwsim) you can do stupid
things like setting up an AP on a certain channel,
then adding another virtual interface and making
that associate on another channel -- this will make
the beaconing to move channel but obviously without
the necessary IEs data update.
In order to improve this situation, first make the
configuration APIs (cfg80211 and nl80211) aware of
multi-channel operation -- we'll eventually need
that in the future anyway. There's one userland API
change and one API addition. The API change is that
now SET_WIPHY must be called with virtual interface
index rather than only wiphy index in order to take
effect for that interface -- luckily all current
users (hostapd) do that. For monitor interfaces, the
old setting is preserved, but monitors are always
slaved to other devices anyway so no guarantees.
The second userland API change is the introduction
of a per virtual interface SET_CHANNEL command, that
hostapd should use going forward to make it easier
to understand what's going on (it can automatically
detect a kernel with this command).
Other than mac80211, no existing cfg80211 drivers
are affected by this change because they only allow
a single virtual interface.
mac80211, however, now needs to be aware that the
channel settings are per interface now, and needs
to disallow (for now) real multi-channel operation,
which is another important part of this patch.
One of the immediate benefits is that you can now
start hostapd to operate on a hardware that already
has a connection on another virtual interface, as
long as you specify the same channel.
Note that two things are left unhandled (this is an
improvement -- not a complete fix):
* different HT/no-HT modes
currently you could start an HT AP and then
connect to a non-HT network on the same channel
which would configure the hardware for no HT;
that can be fixed fairly easily
* CSA
An AP we're connected to on a virtual interface
might indicate switching channels, and in that
case we would follow it, regardless of how many
other interfaces are operating; this requires
more effort to fix but is pretty rare after all
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When reconfiguring an interface due to a previous
hardware restart, mac80211 will currently include
the new IBSS flag on non-IBSS interfaces which may
confuse drivers.
Instead of doing the ~0 trick, simply spell out
which things are going to be reconfigured.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
"mac80211: improve IBSS scanning" was missing a hunk.
This adds that hunk as originally intended.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Currently, hwsim will always detect a double scan
after the first one has finished ...
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When IBSS is fixed to a frequency, it can still
scan to try to find the right BSSID. This makes
sense if the BSSID isn't also fixed, but it need
not scan all channels -- just one is sufficient.
Make it do that by moving the scan setup code to
ieee80211_request_internal_scan() and include
a channel variable setting.
Note that this can be further improved to start
the IBSS right away if both frequency and BSSID
are fixed.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This allows enabling TX and disabling both TX and
RX aggregation sessions manually in debugfs. It is
very useful for debugging session initiation and
teardown problems since with this you don't have
to force a lot of traffic to get aggregation and
thus have less data to analyse.
Also, to debug mac80211 code itself, make hwsim
"support" aggregation sessions. It will still just
transfer the frame, but go through the setup and
teardown handshakes.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Both of these functions can currently return
a station pointer that, to the driver, is
invalid (in IBSS mode only) because adding
the station failed. Check for that, and also
make ieee80211_find_sta() properly use the
per interface station search.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
My previous patch "mac80211: notify driver about
IBSS status" left a problem -- when we merge with
a new BSSID, we never tell the driver that we left
the old one. Fix that.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Some drivers (e.g. iwlwifi) need to know and try
to figure it out based on other things, but making
it explicit is definitely better.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When scanning, it is somewhat important to scan
on the correct virtual interface. All drivers
that currently implement hw_scan only support a
single virtual interface, but that may change
and then we'd want to be ready.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Eric's recent patch added __force, but this
place would seem to require actually doing
a byte order conversion so the printk is
consistent across architectures.
Cc: Eric Dumazet <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
When I updated this from the corresponding
userspace library, an annotation error crept
in -- this variable needs to be annotated as
little endian. No effect on code generation.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
One HT debugging printk is missing a newline,
add it.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Kalle reported that his system deadlocks since my
recent work in this area. The reason quickly became
apparent: we try to cancel_timer_sync() a timer
from within itself. Fix that by making the function
aware of the context it is called from.
Reported-by: Kalle Valo <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Tested-by: Kalle Valo <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When the addba timer expires but has no work to do,
it should not affect the state machine. If it does,
TX will not see the successfully established and we
can also crash trying to re-establish the session.
Cc: [email protected] [2.6.32, 2.6.33]
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Enhance tracing by adding tracing for a variety of
callbacks that the drivers call, and also for
internal calls (currently limited to queue status).
This can aid debugging what is going on in mac80211
in interaction with drivers, since we can now see
what drivers call and not just what mac80211 calls
in the driver.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
A few places in mac80211 do not currently acquire
the sta lock for RX aggregation, but they should.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
The aggregation code has a number of quirks, like
inventing an unneeded WLAN_BACK_TIMER value and
leaking memory under certain circumstances during
station destruction. Fix these issues by using
the regular aggregation session teardown code and
blocking new aggregation sessions, all before the
station is really destructed.
As a side effect, this gets rid of the long code
block to destroy aggregation safely.
Additionally, rename tid_state_rx which can only
have the values IDLE and OPERATIONAL to
tid_active_rx to make it easier to understand
that there is no bitwise stuff going on on the
RX side -- the TX side remains because it needs
to keep track of the driver and peer states.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
I want to use it during station destruction as well
so rename it to WLAN_STA_BLOCK_BA which is also the
only use of it now.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
All callers of ieee80211_sta_stop_rx_ba_session can
just call __ieee80211_stop_rx_ba_session instead
because they already have the station struct, so do
that and remove ieee80211_sta_stop_rx_ba_session.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
All other places except one in the TX path, which
has BHs disabled, and it also cannot be locked from
interrupts so disabling IRQs is not necessary.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When a station w/o a key is destroyed, or when
a driver submits work for a station and thereby
references it again, it seems like potentially
we could reference the station structure while
it is being destroyed.
Wait for an RCU grace period to elapse before
finishing destroying the station after we have
removed the station from the driver and from
the hash table etc., even in the case where no
key is associated with the station.
Also, there's no point in deleting the plink
timer here since it'll be properly deleted just
a bit later.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
The new RCU lockdep support warns about these
in some contexts -- make it aware of the locks
used to protect all this. Different locks are
used in different contexts which unfortunately
means we can't get perfect checking.
Also remove rcu_dereference() from two places
that don't actually dereference the pointers.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Drivers can now advertise to cfg80211 that they have
multiple MAC addresses reserved for a device, but we
don't currently make use of that in mac80211.
Change that and assign different addresses to new
virtual interfaces (if addresses are available) in
order to make it easier for users to use multiple
virtual interfaces; they no longer need to always
assign a new MAC address manually.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
The noise value as is won't be used, isn't
filled by most drivers and doesn't really
make a whole lot of sense on a per packet
basis -- proper cfg80211 survey support in
mac80211 will need to be different.
Mark the struct member as deprecated so it
will be removed from drivers.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
At the wireless summit in Portland we discussed a way of
loading firmware asynchronously from ->probe() before
registration to mac80211, in order to register with the
wireless subsystems with complete information in cases
where firmware is required to know parameters.
This is not yet the case in iwlwifi, but for some new
features we're working on it will be the case since
those will only be supported by new firmware images.
Hence, to start with, convert iwlwifi to load firmware
asynchronously from probe, unbinding the device when
firmware loading fails, and only registering with the
wireless subsystems after firmware has been loaded
successfully.
Future patches will hook into this to register the
new firmware capabilities, depending on the firmware
API version.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Reinette Chatre <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This converts ar9170 to load firmware asynchronously
out of ->probe() and only register with mac80211 when
all firmware has been loaded successfully. If, on the
other hand, any firmware fails to load, it will now
unbind from the device.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This converts mwl8k to use the new station
add/remove callbacks instead of using the
old sta_notify callback.
The new callbacks can sleep, so a lot of
code can be removed now.
Signed-off-by: Johannes Berg <[email protected]>
Acked-by: Lennert Buytenhek <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This converts ath9k to use the new station
add/remove callbacks instead of using the
old sta_notify callback.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This converts mac80211_hwsim to use the new
station add/remove callbacks instead of using
the old sta_notify callback.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This converts p54 to use the new station
add/remove callbacks instead of using the
old sta_notify callback.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This converts ar9170 to use the new station
add/remove callbacks instead of using the
old sta_notify callback.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Currently, ieee80211_drop_unencrypted is called
from management and data frame context, and the
different contexts pass different frames. This
could lead to it processing an 802.3 frame as an
802.11 frame when MFP is enabled.
Move the MFP part of ieee80211_drop_unencrypted
into a new function that is only called for mgmt
frames.
Cc: Jouni Malinen <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
802.11-2007 7.3.1.11 mandates that we need to
reject action frames we don't handle by setting
the 0x80 bit in the category and returning them
to the sender, so do that. In AP mode, hostapd
is responsible for this.
Additionally, drop completely malformed action
frames or ones that should've been encrypted as
unusable, userspace shouldn't see those.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Jouni Malinen <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Fix a copy bug introduced by
commit 47846c9b0c
Author: Johannes Berg <[email protected]>
Date: Wed Nov 25 17:46:19 2009 +0100
mac80211: reduce reliance on netdev
This manifested itself only in debug messages
and in the debugfs rename failure that would
always happen due to trying to rename the dir
over itself.
Signed-off-by: Johannes Berg <[email protected]>
Tested-by: Pavel Roskin <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When userspace requests a deauth while the
authentication work is pending in the auth
(not probe) state, we do not properly abort
the work and then things get confused.
Fix that and also improve the checks here
to include the correct virtual interface,
just in case two virtual interfaces would
ever try to connect to the same BSS.
Also fix a bug -- need to use list_del_rcu
instead of just list_del to free a work
item.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
In AP mode, the only mode where the parameter
is supposed to be valid, we never assign it!
Fix that to allow drivers to avoid parsing
the TIM IE for the value.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Many drivers would like to sleep during station
addition and removal, and currently have a high
complexity there from not being able to.
This introduces two new callbacks sta_add() and
sta_remove() that drivers can implement instead
of using sta_notify() and that can sleep, and
the new sta_add() callback is also allowed to
fail.
The reason we didn't do this previously is that
the IBSS code wants to insert stations from the
RX path, which is a tasklet, so cannot sleep.
This patch will keep the station allocation in
that path, but moves adding the station to the
driver out of line. Since the addition can now
fail, we can have IBSS peer structs the driver
rejected -- in that case we still talk to the
station but never tell the driver about it in
the control.sta pointer. If there will ever be
a driver that has a low limit on the number of
stations and that cannot talk to any stations
that are not known to it, we need to do come up
with a new strategy of handling larger IBSSs,
maybe quicker expiry or rejecting peers.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
We can now easily determine whether we already
have probe response information for the BSS we
are asked to connect to, in which case there's
little point in probing the BSS again.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Upstream radiotap has adopted the namespace
proposal David Young made and I then took care
of, for which I had adapted the radiotap parser
as a library outside the kernel. This brings
the in-kernel parser up to speed.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
For debugging hardware scan trigger/complete
functionality, it was useful to have code in
hwsim that pretends to do a hardware scan.
This code could be extended to actually do the
scan, but for now it was sufficient for me to
only pretend. Since hwsim was written to ease
debugging, it only makes sense to add it to it
permanently.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Reinette found the reason for the warnings that
happened occasionally when a hw-offloaded scan
finished; her description of the problem:
mac80211 will defer the handling of scan requests if it is
busy with management work at the time. The scan requests
are deferred and run after the work has completed. When
this occurs there are currently two problems.
* The scan request for hardware scan is not fully populated
with the band and channels to scan not initialized.
* When the scan is queued the state is not correctly updated
to reflect that a scan is in progress. The problem here is
that when the driver completes the scan and calls
ieee80211_scan_completed() a warning will be triggered
since mac80211 was not aware that a scan was in progress.
The reason is that the queued scan work will start
the hw scan right away when the hw_scan_req struct
has already been allocated. However, in the first
pass it will not have been filled, which happens
at the same time as setting the bits. To fix this,
simply move the allocation after the pending work
test as well, so that the first iteration of the
scan work will call __ieee80211_start_scan() even
in the hardware scan case.
Bug-identified-by: Reinette Chatre <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Because DTIM information is required for powersave
but is only conveyed in beacons, wait for a beacon
before enabling powersave, and change the way the
information is conveyed to the driver accordingly.
mwl8k doesn't currently seem to implement PS but
requires the DTIM period in a different way; after
talking to Lennert we agreed to just have mwl8k do
the parsing itself in the finalize_join work.
Signed-off-by: Johannes Berg <[email protected]>
Acked-by: Lennert Buytenhek <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This new function (previously a static function
called just "find_ie" can be used to find a
specific IE in a buffer of IEs.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Filtered frames not only need their control information
cleared to avoid wrong checks, but also need to have
jiffies and vif assigned so they can be processed or
expired.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Currently, the remain_on_channel work callback needs
to track in its own data structure whether the work
was just started or not. By reordering some code this
becomes unnecessary, the generic wk->started variable
can still be 'false' on the first invocation and only
be 'true' on actual timeout invocations, so that the
extra variable can be removed.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
What a stupid mistake. In
commit 813d766940
Author: Johannes Berg <[email protected]>
Date: Sun Jan 17 01:47:58 2010 +0100
mac80211: move control.hw_key assignment
I inserted code testing the wrong flags field,
which means that the test is almost always true
(it's really testing for the peer's WMM support)
and thus the later parts of the stack assume hw
crypto will be done even if that's not true.
Obviously, that broke software crypto. Maxim
said so specifically, and Jochen probably uses
some cipher that iwl3945 doesn't support in
hardware, which might also explain that Maxim
reports that even hw crypto is broken.
Fix this to test the right flags field.
Reported-by: Maxim Levitsky <[email protected]>
Reported-by: Jochen Friedrich <[email protected]>
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
sparse won't check endianness annotations by
default, but iwlwifi is and should be clean
so we can make sparse check them on it.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Reinette Chatre <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
sparse correctly warns about symbol not
being static, make static to shut it up.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Reinette Chatre <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
The debugfs code can be made a whole lot more
efficient by using debugfs_remove_recursive(),
the large chunk of variables can completely go
away and by moving two variables we no longer
need to allocate an extra chunk of memory.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Reinette Chatre <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
It is better to first notify cfg80211 about the hw
rfkill state (so the rfkill device that will be
registered won't have the wrong state while being
registered), and the power/tt variable init can
(and probably should) also be done first.
Also rename iwl_setup_mac to
iwl_mac_setup_register to better describe what it
really does.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Reinette Chatre <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
It really hasn't been for a long time, not sure
why this stuck around.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Reinette Chatre <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
There are a few station addresses that are
char *, instead of the normal u8 *; gcc
gives pointer signedness warnings for some
of those, so use u8 * consistently.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: Reinette Chatre <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
ieee80211_tx_h_select_key might decide that a frame
need not be encrypted at all, in which case it will
clear tx->key. In that case it may crash if a key
was previously selected, e.g. as the default key.
This is also due to my patch
"mac80211: move control.hw_key assignment".
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
If a device has multiple MAC addresses, userspace will
need to know about that. Similarly, if it allows the
MAC addresses to vary by a bitmask.
If a driver exports multiple addresses, it is assumed
that it will be able to deal with that many different
addresses, which need not necessarily match the ones
programmed into the device; if a mask is set then the
device should deal addresses within that mask based
on an arbitrary "base address".
To test it all and show how it is used, add support
to hwsim even though it can't actually deal with
addresses different from the default.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When a TKIP key is updated, we should pass the station
pointer instead of just the address, since drivers can
use that to store their own data. We also need to pass
the virtual interface pointer.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Future code will need to look up rdev and wdev
within atomic sections, but currently we need
to lock a mutex for such lookups. Change the
list handling for both to be RCU-safe so that
we can look them up in rcu sections instead in
the future.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
In an earlier commit,
mac80211: disable software retry for now
Pavel Roskin reported a problem that seems to be due to
software retry of already transmitted frames. It turns
out that we've never done that correctly, but due to
some recent changes it now crashes in the TX code. I've
added a comment in the patch that explains the problem
better and also points to possible solutions -- which
I can't implement right now.
I disabled software retry of failed/filtered frames
because it was broken. With the work of the previous
patches, it now becomes fairly easy to re-enable it
by adding a flag indicating that the frame shouldn't
be modified, but still running it through the transmit
handlers to populate the control information.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When mac80211 asks a driver to encrypt a frame, it
must assign the control.hw_key pointer for it to
know which key to use etc. Currently, mac80211 does
this whenever it would software-encrypt a frame.
Change the logic of this code to assign the hw_key
pointer when selecting the key, and later check it
when deciding whether to encrypt the frame or let
it be encrypted by the hardware. This allows us to
later simply skip the encryption function since it
no longer modifies the TX control.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
There's no value in setting a flag that will
never be checked after this point, this seems
to be legacy code -- I think previously the
flag was used to check whether to encrypt the
frame or not. Now, however, the flag need not
be set, and setting it actually interferes if
the frame will be processed again later.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
When an skb survived a round-trip through the driver
and needs to be re-used, its control information is
definitely not valid any more, the driver will have
overwritten it.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
This TX handler is used only for assigning the
station pointer in the control information, so
give it a better name. Also move it before rate
control.
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
The type argument to DEBUGFS_ADD() isn't used
and can be removed, it's around from before
the conversion to debugfs_remove_recursive().
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
sparse pointed out that I made a mistake converting
the return value of ieee80211_assoc_success to bool,
this place should return false instead of one of the
enum values (which would be true).
Signed-off-by: Johannes Berg <[email protected]>
Signed-off-by: John W. Linville <[email protected]>
Johannes Berg