[RFC PATCH] brcmfmac: set WIPHY_FLAG_HAVE_AP_SME

[RFC PATCH] brcmfmac: set WIPHY_FLAG_HAVE_AP_SME

[Rafał Miłecki]

This allows user space check NL80211_ATTR_DEVICE_AP_SME and realize it's
dealing with a FullMAC device. It's important for a proper AP setup.
So far user space was only able to guess it's AP_SME after failing to
create monitor interface.

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
---
While this patch makes sense to me, it's breaking user space (hostapd).

There are two similar code paths in hostapd for FullMAC devices. For
details see nl80211_setup_ap, below is my summary:

1) hostapd doesn't get AP_SME
It tries to create monitor interface and after failing to it fallbacks
to the AP_SME:
> nl80211: Setup AP(wlan0) - device_ap_sme=0 use_monitor=1
> Failed to create interface mon.wlan0: -95 (Not supported)
> nl80211: Driver does not support monitor interface type - try to run without it
After that it tries to register for PROBE_REQ management frames but it's
not required to succeed.

2) hostapd knows about AP_SME
It tries to register for ACTION management frames requiring it to
succeed. If it fails it goes like this:
> nl80211: Subscribe to mgmt frames with AP handle 0x99f50 (device SME)
> nl80211: Register frame type=0xd0 (WLAN_FC_STYPE_ACTION) nl_handle=0x99f50 match=
> nl80211: Register frame command failed (type=208): ret=-22 (Invalid argument)
> nl80211: Register frame match - hexdump(len=0): [NULL]
> nl80211: Could not configure driver mode

Now, the problem with brcmfmac is that it doesn't report support for any
management frames in AP mode. It supports sending ACTION (and maybe
PROBE_RESP) frames, but it can't pass ACTION to the cfg80211. So even if
we report we can send ACTION (as it seems we do), it won't fix the
problem.

So this patch will trigger another code path in hostapd that due to
brcmfmac missing support for receiving & passing ACTION will fail.

Do you have any idea how to handle this?

Should hostapd drop depndency on ACTION frames? Should brcmfmac fake
support for receiving & passing ACTION frames? Or should support for
ACTION frames be added indeed?
---
 drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
index 7b01e4d..ab431e5 100644
--- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
+++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
@@ -6180,6 +6180,7 @@ static int brcmf_setup_wiphy(struct wiphy *wiphy, struct brcmf_if *ifp)
 	wiphy->cipher_suites = __wl_cipher_suites;
 	wiphy->n_cipher_suites = ARRAY_SIZE(__wl_cipher_suites);
 	wiphy->flags |= WIPHY_FLAG_PS_ON_BY_DEFAULT |
+			WIPHY_FLAG_HAVE_AP_SME |
 			WIPHY_FLAG_OFFCHAN_TX |
 			WIPHY_FLAG_HAS_REMAIN_ON_CHANNEL;
 	if (brcmf_feat_is_enabled(ifp, BRCMF_FEAT_TDLS))
-- 
1.8.4.5

Re: [RFC PATCH] brcmfmac: set WIPHY_FLAG_HAVE_AP_SME

[Arend van Spriel]

On 20-1-2016 13:47, Rafał Miłecki wrote:
> This allows user space check NL80211_ATTR_DEVICE_AP_SME and realize it's
> dealing with a FullMAC device. It's important for a proper AP setup.
> So far user space was only able to guess it's AP_SME after failing to
> create monitor interface.
> 
> Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
> ---
> While this patch makes sense to me, it's breaking user space (hostapd).
> 
> There are two similar code paths in hostapd for FullMAC devices. For
> details see nl80211_setup_ap, below is my summary:
> 
> 1) hostapd doesn't get AP_SME
> It tries to create monitor interface and after failing to it fallbacks
> to the AP_SME:
>> nl80211: Setup AP(wlan0) - device_ap_sme=0 use_monitor=1
>> Failed to create interface mon.wlan0: -95 (Not supported)
>> nl80211: Driver does not support monitor interface type - try to run without it
> After that it tries to register for PROBE_REQ management frames but it's
> not required to succeed.
> 
> 2) hostapd knows about AP_SME
> It tries to register for ACTION management frames requiring it to
> succeed. If it fails it goes like this:
>> nl80211: Subscribe to mgmt frames with AP handle 0x99f50 (device SME)
>> nl80211: Register frame type=0xd0 (WLAN_FC_STYPE_ACTION) nl_handle=0x99f50 match=
>> nl80211: Register frame command failed (type=208): ret=-22 (Invalid argument)
>> nl80211: Register frame match - hexdump(len=0): [NULL]
>> nl80211: Could not configure driver mode
> 
> Now, the problem with brcmfmac is that it doesn't report support for any
> management frames in AP mode. It supports sending ACTION (and maybe
> PROBE_RESP) frames, but it can't pass ACTION to the cfg80211. So even if
> we report we can send ACTION (as it seems we do), it won't fix the
> problem.
> 
> So this patch will trigger another code path in hostapd that due to
> brcmfmac missing support for receiving & passing ACTION will fail.
> 
> Do you have any idea how to handle this?

The mgmt_tx callback in brcmfmac was added for P2P functionality. So if
anything, a check should be added to assure it is only used on P2P
interfaces.

> Should hostapd drop depndency on ACTION frames? Should brcmfmac fake
> support for receiving & passing ACTION frames? Or should support for
> ACTION frames be added indeed?

So why not stick with the current implementation. It may not seem
optimal to first try one thing and fallback to another, but at least it
works. I was thinking about trying with use_monitor=0, but looking in
the code nl80211_setup_ap() that will not resolve it either.

Regards,
Arend

> ---
>  drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> index 7b01e4d..ab431e5 100644
> --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
> @@ -6180,6 +6180,7 @@ static int brcmf_setup_wiphy(struct wiphy *wiphy, struct brcmf_if *ifp)
>  	wiphy->cipher_suites = __wl_cipher_suites;
>  	wiphy->n_cipher_suites = ARRAY_SIZE(__wl_cipher_suites);
>  	wiphy->flags |= WIPHY_FLAG_PS_ON_BY_DEFAULT |
> +			WIPHY_FLAG_HAVE_AP_SME |
>  			WIPHY_FLAG_OFFCHAN_TX |
>  			WIPHY_FLAG_HAS_REMAIN_ON_CHANNEL;
>  	if (brcmf_feat_is_enabled(ifp, BRCMF_FEAT_TDLS))
> 

Re: [RFC PATCH] brcmfmac: set WIPHY_FLAG_HAVE_AP_SME

[Rafał Miłecki]

On 25 January 2016 at 10:49, Arend van Spriel <arend@broadcom.com> wrote:
> On 20-1-2016 13:47, Rafał Miłecki wrote:
>> Now, the problem with brcmfmac is that it doesn't report support for any
>> management frames in AP mode. It supports sending ACTION (and maybe
>> PROBE_RESP) frames, but it can't pass ACTION to the cfg80211. So even if
>> we report we can send ACTION (as it seems we do), it won't fix the
>> problem.
>>
>> So this patch will trigger another code path in hostapd that due to
>> brcmfmac missing support for receiving & passing ACTION will fail.
>>
>> Do you have any idea how to handle this?
>
> The mgmt_tx callback in brcmfmac was added for P2P functionality. So if
> anything, a check should be added to assure it is only used on P2P
> interfaces.
>
>> Should hostapd drop depndency on ACTION frames? Should brcmfmac fake
>> support for receiving & passing ACTION frames? Or should support for
>> ACTION frames be added indeed?
>
> So why not stick with the current implementation. It may not seem
> optimal to first try one thing and fallback to another, but at least it
> works. I was thinking about trying with use_monitor=0, but looking in
> the code nl80211_setup_ap() that will not resolve it either.

I don't like current implementation as it depends on some failure,
fallback & inconsistent code in hostapd. It seems hacky to me and I
can imagine things going wrong on some simple hostapd change. I beliee
it also doesn't allow brcmfmac to implement monitor mode, because once
it's done, hostapd won't hit the same fallback path.