pre-merge issues

mac80211

Make the stack SMP safe.
hide (or get rid of) master dev to have stable userland interface
Review locking around passive scanning (ieee80211_rx_h_passive_scan)
if there is absolutely no possibility to retain the current wlan%d interface names, change default interface name to eth%d
fix per-netdev debugfs stuff (needs debugfs_rename)
remove ieee80211_netif_oper

cfg80211

propagate phy renaming to debugfs (needs debugfs_rename)

cfg80211

rename struct wiphy to struct phy80211?

mac80211

Most important issues

Add callbacks for devices with advanced firmware.
Need to stop TX/RX when a radar is detected for the duration of scan for a new channel. (Partially solved by PRISM2_PARAM_RADIO_ENABLED.) [reported by Jouni Malinen]
get rid of master dev
rename ieee80211_local (e.g.: rename ieee80211_hw to ieee80211_dev and ieee80211_local to ieee80211_hw)
don't allow removal of qdisc
compat: handle reassociation when the sta is out of range for a moment and then comes back

Other issues

Add a ieee80211_resume(hw) function that drivers can call when they resume. It should
- call set_key for each key that was uploaded to hardware
- reconfigure hardware (channel etc)
- ...?

crypto

Embed ieee80211_key_conf into ieee80211_key so just one memcpy can be used in ieee80211_key_data2conf().
ieee80211_rx_michael_mic_report: comparing to IEEE80211_IF_TYPE_AP yields to false everytime. (FIXME: haven't it be fixed already?)
split out frame crypto stuff into modules like in-kernel ieee80211 does. I'm not sure any more this is worth it though since the actual algorithms (i.e. most code) can be modular anyway – JohannesBerg DateTime(2007-04-06T09:48:48Z)
Calls to ieee80211_key_threshold_notify are needless in some places. When the interface is in STA mode, a lot of operations is performed and the result is thrown away.

scanning

Merge both types of scanning: software one (ieee80211_sta_scan_timer) and hardware/hostapd one (ieee80211_scan_handler).
Rework background scanning to work without background scan callback. [suggested by Johannes Berg]

core

Using of radiotap instead of Prism monitoring header.
Add dormant state support.
"Strawman proposal for a generic frame format" (http://linux-net.osdl.org/index.php/Wireless:Txformat) [suggested by Solomon Peachy]
RCPI support (http://www.spinics.net/lists/netdev/msg12028.html) [suggested by Simon Barber]
ieee80211_regdom and ieee80211_japan_5ghz module paramaters should go away in favour of userspace regulatory app.
Add support for multiple queue netdevices (if Peter Waskiewicz's patch is accepted).

misc

Export information about softmac/fullmac type of the hardware to userspace. [suggested by Jouni Malinen] (maybe export capabilities instead?)
Easier handling of configuration requests (ieee80211_hw->config()). In the current implementation, driver has no way to find out which parameter is being set; setting all parameters on every call to config() is obviously not a good idea.
struct ieee80211_tx_queue_stats is strange (why not to use ieee80211_tx_queue_stats_data directly?)
Kicking DMA on the last fragment only (?) [suggested by Ivo van Doorn]
Why there is a beacon_update() handler? Maybe it would be easier for the driver to be allowed to call ieee80211_beacon_get() even in IBSS mode.
Add functions that will tell the driver which rates and modulations are allowed. Some cards need to tell their firmware about it.
Move ETH_P_PAE from ieee80211_i.h to linux/if_ether.h.
Make IEEE80211_FRAGMENT_MAX configurable (preferably at run-time).
Alter LL_MAX_HEADER constant.
Fix the race in PS status update (see TODO in ieee80211_tx_status()).
Do not misuse iwpriv for generic ioctls (i.e. do not use iwpriv ioctls in the stack at all).
Remove ioctl (add_iface, WPA) compatibility stuff.
Move ieee80211_get_hdrlen* from ieee80211.c to ieee80211_proto.c.
Sort out function definitions in ieee80211_i.h (they are at two different places in the file now).
Split ieee80211.c into several files.
Add spin_lock_bh(&local->sub_if_lock) around accesses to local->sub_if_list.
ieee80211_if_conf should be a part of ieee80211_sub_if_data
In case of STA or IBSS, after a change of SSID or generic_element, ieee80211_if_conf should be called.
ieee80211_ioctl_remove_sta: sta_info_free should be called with locked = 0
Clean up headers mess (http://www.spinics.net/lists/netdev/msg12124.html). [suggested by Johannes Berg]
Get rid of bitfields (partially done, still left: ieee80211_txrx_data, ieee80211_tx_packet_data, ieee80211_tx_stored_packet, ieee80211_passive_scan, ieee80211_if_sta and ieee80211_sub_if_data). [suggested by Jeff Garzik]
#ifdef out wpa_test variable (but do not add #ifdefs into ifs).
Allow set_key to return -ENOBUFS to indicate there is no more room for keys and do sw encryption for this STA then. [suggested by Michael Buesch]
ieee80211_msg_type, ieee80211_phytype and ieee80211_ssi_type should be in capitals (I think they can be removed once we move to radiotap)
Allow "even-harder-monitor" mode - i.e. all frames (including those with a bad FCS) are received. That means introducing a new "not_checksummed" flag for non-monitor interfaces to operate simultaneously.
For research: do the same with incorrect PLCP checksums?

qdisc bugs

From Patrick McHardy:

wme_qdiscop_enqueue doesn't increment q.qlen for packets queued to q->requeued[], which might cause upper layer code to stop dequeueing if q.qlen reaches zero.
classify_1d doesn't care about tc_classify return values. tc_classify may decide to steal packets, drop them, etc. In case of stolen packets this causes use-after-free, otherwise just malfunctions.
classify_1d returns res.class if it is != -1, which can never happen (except with an empty classifier list because of the explicit initialization, but you should check the return code) since ->get() and ->bind_tcf() both return 0 for invalid classes and the classid otherwise. There's also an off-by-one, classids start at one, so it should return res.class - 1 (or better res.classid - 1, which is meant to be a numerical identifier).
wme_discop_destroy leaks classifier module references and memory when destroying classifiers, it should use tcf_destroy()

Considering that it is possibly and may be desirable to attach a different qdisc than the built-in multiband qdisc, it might also make sense to split the 80211 specific classification in a seperate classifier module to allow simple classification of management traffic with other qdiscs.

Library functions we should add

PLCP Length calculation
Functions to fill modes/rates for B, BG, and ABG.
TKIP key mixing functions (work in progress by mbuesch)

Optimizations

ieee80211_get_hdrlen and ieee80211_is_eapol are called very often.
Inline ieee80211_get_bssid()?
All frames go through all of decryption functions. It should be possible to know what to expect and only pass it through those functions that are needed. [suggested by Johannes Berg]
Recognition of device incoming frame belongs to can be made much smarter and faster.
When one packet is dumped through several interfaces, some operations can be performed just once (e.g. searching for the key, sometimes decrypting, defragmentation etc.).
Add #ifdef's for not compiling AP stuff. [suggested by Jouni Malinen]
Move fragmentation etc. into 802.11 qdisc. [suggested by Simon Barber]
Try to switch from sw crypto to hw when there was no more room for STA keys in the hardware and one STA disassociates (so there is possibly a room now). [suggested by Johannes and Michael]
dynamically registered tx/rx handlers (and split them out from ieee80211.c)

Coding style

remove forward declarations