The mac80211 softmac driver subsystem and handling of monitor interfaces

The mac80211 softmac driver subsystem and handling of monitor interfaces

[Roger James]

Hello,

For various reasons I have been looking into the handling of monitor 
virtual interfaces in drivers that link with the mac80211 module.
It is my understanding that monitor interfaces are not "notified" to the 
driver module, and that driver modules should handle the 
IEEE80211_CONF_CHANGE_MONITOR flag passed down by the config callback 
whenever the set of virtual interfaces changes between having zero and 
one monitor member. I am basing this assumption on Johannes's 2009 
presentation

    "relevancy to drivers
\x0f
    drivers need to allow each interface type
\x0f
    drivers need to support certain operations for certain interface
    types
\x0f
    drivers can support multiple virtual interfaces
\x0f
    but:
    drivers not notified of monitor interface"

and the somewhat opaque comments in mac80211.h

    "@IEEE80211_CONF_MONITOR: there's a monitor interface present --
     use this to determine for example whether to calculate timestamps
     for packets or not, do not use instead of filter flags!"

Firstly is this assumption correct?

Looking at the drivers in the current mainline kernel that link the 
mac80211 modules only a small minority of them hook this event. They are :-

ath/ath10k ath/ath9k (both ath9k and ath9k_htc) cw1200 brcm80211/brcmsmac

All the others seem to be doing things with NL80211_IFTYPE_MONITOR in 
the vif structure. This seems contradictory.

If my assumptions are true I would expect drivers to do whatever private 
interaction with the hardware to enable monitoring at the point
they receive the monitor change notification.  For example setting up 
the right frame headers and frame types to pass up etc. I wonder why the 
majority of drivers seem to be doing it elsewhere. I assume this is 
historical. I also assume that hw filters are handled separately so that 
different monitor vifs can have different flag sets.

Please feel free to tear this to shreds! I am on a steep learning curve 
on this. I have read most of the docs I could find and studied the code, 
but please let me have any suggestions you have for further research.

For information I have appended a couple of greps in the current 
mainline kernel tree and a list of modules in my current running kernel 
linking mac80211, after the end of this note.

Thanks
Roger

roger@dragon:~/linux-mainline/drivers/net/wireless$ find . -name 
"*.[ch]" -exec rgrep NL80211_IFTYPE_MONITOR {} \; -print
		{NL80211_IFTYPE_MONITOR,	WMI_NETTYPE_ADHOC}, /* FIXME */
	case NL80211_IFTYPE_MONITOR:
				 BIT(NL80211_IFTYPE_MONITOR);
./ath/wil6210/cfg80211.c
	if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) {
./ath/wil6210/ethtool.c
	if (wdev->iftype == NL80211_IFTYPE_MONITOR) {
./ath/wil6210/wmi.c
		if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) {
	if (unlikely(wil->wdev->iftype == NL80211_IFTYPE_MONITOR)) {
./ath/wil6210/txrx.c
	if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR)
./ath/wil6210/interrupt.c
	case NL80211_IFTYPE_MONITOR:
./ath/wil6210/pm.c
	case NL80211_IFTYPE_MONITOR:
	case NL80211_IFTYPE_MONITOR:
./ath/wil6210/main.c
	case NL80211_IFTYPE_MONITOR:
./ath/debug.c
	if (!vif || vif->type == NL80211_IFTYPE_MONITOR)
	case NL80211_IFTYPE_MONITOR:
	if (vif->type == NL80211_IFTYPE_MONITOR) {
	if (vif->type == NL80211_IFTYPE_MONITOR) {
	if (vif->type == NL80211_IFTYPE_MONITOR) {
	if (vif->type == NL80211_IFTYPE_MONITOR) {
./ath/ath10k/mac.c
		priv->ah->opmode = NL80211_IFTYPE_MONITOR;
./ath/ath9k/htc_drv_main.c
	case NL80211_IFTYPE_MONITOR:
	case NL80211_IFTYPE_MONITOR:
./ath/ath5k/pcu.c
		     ah->opmode == NL80211_IFTYPE_MONITOR)) {
./ath/ath5k/base.c
	case NL80211_IFTYPE_MONITOR:
./zydas/zd1211rw/zd_mac.c
	case NL80211_IFTYPE_MONITOR:
./st/cw1200/debug.c
	priv->mode = NL80211_IFTYPE_MONITOR;
	if (priv->mode != NL80211_IFTYPE_MONITOR) {
	priv->mode = NL80211_IFTYPE_MONITOR;
	case NL80211_IFTYPE_MONITOR:
	    priv->mode == NL80211_IFTYPE_MONITOR ||
./st/cw1200/sta.c
	case NL80211_IFTYPE_MONITOR:
./st/cw1200/wsm.c
	case NL80211_IFTYPE_MONITOR:
	if (vif->type == NL80211_IFTYPE_MONITOR) {
	case NL80211_IFTYPE_MONITOR:
./intel/iwlwifi/mvm/mac80211.c
	case NL80211_IFTYPE_MONITOR:
./intel/iwlwifi/mvm/quota.c
	case NL80211_IFTYPE_MONITOR:
	WARN_ON(vif->type != NL80211_IFTYPE_MONITOR);
	case NL80211_IFTYPE_MONITOR:
	if (vif->type == NL80211_IFTYPE_MONITOR) {
./intel/iwlwifi/mvm/mac-ctxt.c
	case NL80211_IFTYPE_MONITOR:
./intel/iwlwifi/mvm/power.c
	if (vif->type == NL80211_IFTYPE_MONITOR)
./intel/iwlwifi/dvm/mac80211.c
	case NL80211_IFTYPE_MONITOR:
./intel/iwlwifi/dvm/rxon.c
		BIT(NL80211_IFTYPE_ADHOC) | BIT(NL80211_IFTYPE_MONITOR);
./intel/iwlwifi/dvm/main.c
	case NL80211_IFTYPE_MONITOR:
		wdev->wiphy->interface_modes |= BIT(NL80211_IFTYPE_MONITOR);
./marvell/libertas/cfg.c
	if (priv->wdev->iftype == NL80211_IFTYPE_MONITOR) {
./marvell/libertas/mesh.c
	if (priv->wdev->iftype == NL80211_IFTYPE_MONITOR) {
	if (priv->wdev->iftype == NL80211_IFTYPE_MONITOR) {
	if (priv->wdev->iftype != NL80211_IFTYPE_MONITOR ||
./marvell/libertas/tx.c
	case NL80211_IFTYPE_MONITOR:
		if (priv->wdev->iftype == NL80211_IFTYPE_MONITOR)
		if (priv->wdev->iftype == NL80211_IFTYPE_MONITOR)
./marvell/libertas/main.c
	if (priv->wdev->iftype == NL80211_IFTYPE_MONITOR) {
./marvell/libertas/rx.c
			if (priv->wdev->iftype == NL80211_IFTYPE_MONITOR &&
./marvell/libertas/cmdresp.c
	case NL80211_IFTYPE_MONITOR:
	case NL80211_IFTYPE_MONITOR:
	case NL80211_IFTYPE_MONITOR:
./broadcom/brcm80211/brcmfmac/cfg80211.c
		case NL80211_IFTYPE_MONITOR:
./intersil/p54/fwio.c
	case NL80211_IFTYPE_MONITOR:
./intersil/p54/txrx.c
	priv->mode = NL80211_IFTYPE_MONITOR;
	if (priv->mode != NL80211_IFTYPE_MONITOR) {
	priv->mode = NL80211_IFTYPE_MONITOR;
./intersil/p54/main.c
		wiphy->interface_modes |= BIT(NL80211_IFTYPE_MONITOR);
	case NL80211_IFTYPE_MONITOR:
	if (priv->iw_mode == NL80211_IFTYPE_MONITOR) {
./intersil/orinoco/cfg.c
	if (priv->iw_mode == NL80211_IFTYPE_MONITOR) {
./intersil/orinoco/wext.c
	    (priv->iw_mode == NL80211_IFTYPE_MONITOR)) {
./intersil/orinoco/orinoco_usb.c
	if (priv->iw_mode == NL80211_IFTYPE_MONITOR) {
		if (priv->iw_mode == NL80211_IFTYPE_MONITOR)
	if (priv->iw_mode == NL80211_IFTYPE_MONITOR) {
./intersil/orinoco/hw.c
	case NL80211_IFTYPE_MONITOR:
	    (priv->iw_mode == NL80211_IFTYPE_MONITOR)) {
	if (priv->iw_mode == NL80211_IFTYPE_MONITOR) {
		if (priv->iw_mode == NL80211_IFTYPE_MONITOR)
./intersil/orinoco/main.c
	case NL80211_IFTYPE_MONITOR:
	if (priv->mode != NL80211_IFTYPE_MONITOR)
	priv->mode = NL80211_IFTYPE_MONITOR;
	priv->mode = NL80211_IFTYPE_MONITOR;
./admtek/adm8211.c
roger@dragon:~/linux-mainline/drivers/net/wireless$ find . -name 
"*.[ch]" -exec rgrep IEEE80211_CONF_CHANGE_MONITOR {} \; -print
	if (changed & IEEE80211_CONF_CHANGE_MONITOR) {
./ath/ath10k/mac.c
	if (changed & IEEE80211_CONF_CHANGE_MONITOR) {
./ath/ath9k/htc_drv_main.c
	if (changed & IEEE80211_CONF_CHANGE_MONITOR) {
./ath/ath9k/main.c
	if (changed & IEEE80211_CONF_CHANGE_MONITOR) {
./ti/wl1251/main.c
	if (changed & IEEE80211_CONF_CHANGE_MONITOR) {
./st/cw1200/sta.c
	if (changed & IEEE80211_CONF_CHANGE_MONITOR)
./broadcom/brcm80211/brcmsmac/mac80211_if.c

The following comes from the kernel I am currently running (4.2.0) not 
the latest mainline.

roger@dragon:/lib/modules/4.2.0-30-generic$ grep mac80211 modules.dep | 
cut -d: -f1
kernel/drivers/net/wireless/mediatek/mt7601u/mt7601u.ko
kernel/drivers/net/wireless/ti/wlcore/wlcore.ko
kernel/drivers/net/wireless/ti/wl12xx/wl12xx.ko
kernel/drivers/net/wireless/ti/wl1251/wl1251.ko
kernel/drivers/net/wireless/ti/wl1251/wl1251_spi.ko
kernel/drivers/net/wireless/ti/wl1251/wl1251_sdio.ko
kernel/drivers/net/wireless/ti/wl18xx/wl18xx.ko
kernel/drivers/net/wireless/at76c50x-usb.ko
kernel/drivers/net/wireless/b43/b43.ko
kernel/drivers/net/wireless/b43legacy/b43legacy.ko
kernel/drivers/net/wireless/zd1211rw/zd1211rw.ko
kernel/drivers/net/wireless/rtl818x/rtl8180/rtl818x_pci.ko
kernel/drivers/net/wireless/rtl818x/rtl8187/rtl8187.ko
kernel/drivers/net/wireless/rtlwifi/rtlwifi.ko
kernel/drivers/net/wireless/rtlwifi/rtl_pci.ko
kernel/drivers/net/wireless/rtlwifi/rtl_usb.ko
kernel/drivers/net/wireless/rtlwifi/rtl8192c/rtl8192c-common.ko
kernel/drivers/net/wireless/rtlwifi/rtl8192ce/rtl8192ce.ko
kernel/drivers/net/wireless/rtlwifi/rtl8192cu/rtl8192cu.ko
kernel/drivers/net/wireless/rtlwifi/rtl8192se/rtl8192se.ko
kernel/drivers/net/wireless/rtlwifi/rtl8192de/rtl8192de.ko
kernel/drivers/net/wireless/rtlwifi/rtl8723ae/rtl8723ae.ko
kernel/drivers/net/wireless/rtlwifi/rtl8723be/rtl8723be.ko
kernel/drivers/net/wireless/rtlwifi/rtl8188ee/rtl8188ee.ko
kernel/drivers/net/wireless/rtlwifi/rtl8821ae/rtl8821ae.ko
kernel/drivers/net/wireless/rtlwifi/rtl8192ee/rtl8192ee.ko
kernel/drivers/net/wireless/libertas_tf/libertas_tf.ko
kernel/drivers/net/wireless/libertas_tf/libertas_tf_usb.ko
kernel/drivers/net/wireless/adm8211.ko
kernel/drivers/net/wireless/mwl8k.ko
kernel/drivers/net/wireless/iwlwifi/dvm/iwldvm.ko
kernel/drivers/net/wireless/iwlwifi/mvm/iwlmvm.ko
kernel/drivers/net/wireless/iwlegacy/iwlegacy.ko
kernel/drivers/net/wireless/iwlegacy/iwl4965.ko
kernel/drivers/net/wireless/iwlegacy/iwl3945.ko
kernel/drivers/net/wireless/rt2x00/rt2x00lib.ko
kernel/drivers/net/wireless/rt2x00/rt2x00mmio.ko
kernel/drivers/net/wireless/rt2x00/rt2x00pci.ko
kernel/drivers/net/wireless/rt2x00/rt2x00usb.ko
kernel/drivers/net/wireless/rt2x00/rt2800lib.ko
kernel/drivers/net/wireless/rt2x00/rt2800mmio.ko
kernel/drivers/net/wireless/rt2x00/rt2400pci.ko
kernel/drivers/net/wireless/rt2x00/rt2500pci.ko
kernel/drivers/net/wireless/rt2x00/rt61pci.ko
kernel/drivers/net/wireless/rt2x00/rt2800pci.ko
kernel/drivers/net/wireless/rt2x00/rt2500usb.ko
kernel/drivers/net/wireless/rt2x00/rt73usb.ko
kernel/drivers/net/wireless/rt2x00/rt2800usb.ko
kernel/drivers/net/wireless/p54/p54common.ko
kernel/drivers/net/wireless/p54/p54usb.ko
kernel/drivers/net/wireless/p54/p54pci.ko
kernel/drivers/net/wireless/p54/p54spi.ko
kernel/drivers/net/wireless/ath/ath5k/ath5k.ko
kernel/drivers/net/wireless/ath/ath9k/ath9k.ko
kernel/drivers/net/wireless/ath/ath9k/ath9k_htc.ko
kernel/drivers/net/wireless/ath/carl9170/carl9170.ko
kernel/drivers/net/wireless/ath/ar5523/ar5523.ko
kernel/drivers/net/wireless/ath/ath10k/ath10k_core.ko
kernel/drivers/net/wireless/ath/ath10k/ath10k_pci.ko
kernel/drivers/net/wireless/ath/wcn36xx/wcn36xx.ko
kernel/drivers/net/wireless/mac80211_hwsim.ko
kernel/drivers/net/wireless/brcm80211/brcmsmac/brcmsmac.ko
kernel/drivers/net/wireless/cw1200/cw1200_core.ko
kernel/drivers/net/wireless/cw1200/cw1200_wlan_sdio.ko
kernel/drivers/net/wireless/cw1200/cw1200_wlan_spi.ko
kernel/drivers/net/wireless/rsi/rsi_91x.ko
kernel/drivers/net/wireless/rsi/rsi_sdio.ko
kernel/drivers/net/wireless/rsi/rsi_usb.ko
kernel/drivers/staging/vt6655/vt6655_stage.ko
kernel/drivers/staging/vt6656/vt6656_stage.ko
kernel/net/mac80211/mac80211.ko

Re: The mac80211 softmac driver subsystem and handling of monitor interfaces

[Johannes Berg]

On Tue, 2016-03-15 at 10:42 +0000, Roger James wrote:

> ath/ath10k ath/ath9k (both ath9k and ath9k_htc) cw1200
> brcm80211/brcmsmac
> 
> All the others seem to be doing things with NL80211_IFTYPE_MONITOR in
> the vif structure. This seems contradictory.
> 

You should look into the WANT_MONITOR flag. :)

johannes

Re: The mac80211 softmac driver subsystem and handling of monitor interfaces

[Roger James]

On 15/03/16 12:15, Johannes Berg wrote:
> On Tue, 2016-03-15 at 10:42 +0000, Roger James wrote:
>
>> ath/ath10k ath/ath9k (both ath9k and ath9k_htc) cw1200
>> brcm80211/brcmsmac
>>
>> All the others seem to be doing things with NL80211_IFTYPE_MONITOR in
>> the vif structure. This seems contradictory.
>>
> You should look into the WANT_MONITOR flag. :)
>
> johannes
> --
>
Hmm...

roger@dragon:~/linux-mainline$ find . -name "*.[ch]" -exec grep -n 
IEEE80211_HW_WANT_MONITOR_VIF {} \; -print
1851: * @IEEE80211_HW_WANT_MONITOR_VIF: The driver would like to be 
informed of
1928:    IEEE80211_HW_WANT_MONITOR_VIF,
./include/net/mac80211.h

Is that what you meant. Nobody seems to be using it. Even mac80211 
itself. Or am I being stupid again :-)



Roger

Re: The mac80211 softmac driver subsystem and handling of monitor interfaces

[Johannes Berg]

On Tue, 2016-03-15 at 13:01 +0000, Roger James wrote:
> 
> roger@dragon:~/linux-mainline$ find . -name "*.[ch]" -exec grep -n 
> IEEE80211_HW_WANT_MONITOR_VIF {} \; -print
> 1851: * @IEEE80211_HW_WANT_MONITOR_VIF: The driver would like to be 
> informed of
> 1928:    IEEE80211_HW_WANT_MONITOR_VIF,
> ./include/net/mac80211.h
> 
> Is that what you meant. Nobody seems to be using it. Even mac80211 
> itself. Or am I being stupid again :-)
> 

There are macros generating the checks and setting it, so you want to
grep without the IEEE80211_HW_ prefix

johannes

Re: The mac80211 softmac driver subsystem and handling of monitor interfaces

[Roger James]

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

Re: The mac80211 softmac driver subsystem and handling of monitor interfaces

[Johannes Berg]

On Wed, 2016-03-16 at 09:53 +0000, Roger James wrote:
> 
> However that only accounts for the ath10k, iwldvm, and iwlmvm
> drivers. I realise that there is a lot of history here, but is what
> the remaining drivers doing in any way deprecated? Also can anyone
> give me a heads up on what the architectural difference is between
> hooking the monitor config change and and setting the want monitor
> flag. Is one way preferable to the other? Both seem to inform the
> driver that there is a transition between having no monitor virtual
> interfaces and having one or more.

Not quite. CONF_MONITOR says "you have >0" monitor interfaces.
drv_add_interface(type=monitor) says "you have >0 monitor interfaces
and 0 others", so there's a difference.

Checking type==monitor in add_interface() *without*
setting WANT_MONITOR_VIF will never be true, since that cannot happen.
Not sure which drivers are doing that, and why they'd try to.

johannes

Re: The mac80211 softmac driver subsystem and handling of monitor interfaces

[Roger James]

On 15 Mar 2016 13:08, Johannes Berg <johannes@sipsolutions.net> wrote:
 >
 > On Tue, 2016-03-15 at 13:01 +0000, Roger James wrote:
 > >
 > > roger@dragon:~/linux-mainline$ find . -name "*.[ch]" -exec grep -n
 > > IEEE80211_HW_WANT_MONITOR_VIF {} \; -print
 > > 1851: * @IEEE80211_HW_WANT_MONITOR_VIF: The driver would like to be
 > > informed of
 > > 1928:    IEEE80211_HW_WANT_MONITOR_VIF,
 > > ./include/net/mac80211.h
 > >
 > > Is that what you meant. Nobody seems to be using it. Even mac80211
 > > itself. Or am I being stupid again :-)
 > >
 >
 > There are macros generating the checks and setting it, so you want to
 > grep without the IEEE80211_HW_ prefix
 >
 > johannes
 > --


Thanks guys I have got it now.

However that only accounts for the ath10k, iwldvm, and iwlmvm drivers. I 
realise that there is a lot of history here, but is what the remaining 
drivers doing in any way deprecated? Also can anyone give me a heads up 
on what the architectural difference is between hooking the monitor 
config change and and setting the want monitor flag. Is one way 
preferable to the other? Both seem to inform the driver that there is a 
transition between having no monitor virtual interfaces and having one 
or more.

Cheers,

Roger