Hi,

> -----Original Message-----
> From: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
> Sent: Friday, January 28, 2022 5:53 AM
> To: Pkshih <pkshih@realtek.com>
> Cc: linux-wireless@vger.kernel.org; tony0620emma@gmail.com; kvalo@codeaurora.org;
> johannes@sipsolutions.net; netdev@vger.kernel.org; linux-kernel@vger.kernel.org; Neo Jou
> <neojou@gmail.com>; Jernej Skrabec <jernej.skrabec@gmail.com>; Ed Swierk <eswierk@gh.st>
> Subject: Re: [PATCH v3 0/8] rtw88: prepare locking for SDIO support
> 
> Hi Ping-Ke,
> 
> On Mon, Jan 24, 2022 at 3:59 AM Pkshih <pkshih@realtek.com> wrote:

[...]

> >
> > To avoid this, we can add a flag to struct rtw_vif, and set this flag
> > when ::remove_interface. Then, only collect vif without this flag into list
> > when we use iterate_actiom().
> >
> > As well as ieee80211_sta can do similar fix.
> >

I would prefer my method that adds a 'bool disabled' flag to struct rtw_vif/rtw_sta
and set it when ::remove_interface/::sta_remove. Then rtw_iterate_stas() can
check this flag to decide whether does thing or not.

[...]

> 
> For the sta use-case I thought about adding a dedicated rwlock
> (include/linux/rwlock.h) for rtw_dev->mac_id_map.
> rtw_sta_{add,remove} would take a write-lock.
> rtw_iterate_stas() takes the read-lock (the lock would be acquired
> before calling into ieee80211_iterate_...). Additionally
> rtw_iterate_stas() needs to check if the station is still valid
> according to mac_id_map - if not: skip/ignore it for that iteration.
> This could be combined with your
> 0001-rtw88-use-atomic-to-collect-stas-and-does-iterators.patch.

Using a 'disabled' flag within rtw_vif/rtw_sta will be intuitive and
better than bitmap of mac_id_map. Please reference my mention above.

> 
> For the interface use-case it's not clear to me how this works at all.
> rtw_ops_add_interface() has (in a simplified view):
>     u8 port = 0;
>     // the port variable is never changed
>     rtwvif->port = port;
>     rtwvif->conf = &rtw_vif_port[port];
>     rtw_info(rtwdev, "start vif %pM on port %d\n", vif->addr, rtwvif->port);
> How do multiple interfaces (vifs) work in rtw88 if the port is always
> zero? Is some kind of tracking of the used ports missing (similar to
> how we track the used station IDs - also called mac_id - in
> rtw_dev->mac_id_map)?

The port should be allocated dynamically if we support two or more vifs.
We have internal tree that is going to support p2p by second vif.


Ping-Ke

Hi Ping-Ke,

On Fri, Jan 28, 2022 at 1:51 AM Pkshih <pkshih@realtek.com> wrote:
[...]
>
> > >
> > > To avoid this, we can add a flag to struct rtw_vif, and set this flag
> > > when ::remove_interface. Then, only collect vif without this flag into list
> > > when we use iterate_actiom().
> > >
> > > As well as ieee80211_sta can do similar fix.
> > >
>
> I would prefer my method that adds a 'bool disabled' flag to struct rtw_vif/rtw_sta
> and set it when ::remove_interface/::sta_remove. Then rtw_iterate_stas() can
> check this flag to decide whether does thing or not.
That would indeed be a very straight forward approach and easy to read.
In net/mac80211/iface.c there's some cases where after
drv_remove_interface() (which internally calls our .remove_interface
op) will kfree the vif (sdata). Doesn't that then result in a
use-after-free if we rely on a boolean within rtw_vif?

[...]
> > For the interface use-case it's not clear to me how this works at all.
> > rtw_ops_add_interface() has (in a simplified view):
> >     u8 port = 0;
> >     // the port variable is never changed
> >     rtwvif->port = port;
> >     rtwvif->conf = &rtw_vif_port[port];
> >     rtw_info(rtwdev, "start vif %pM on port %d\n", vif->addr, rtwvif->port);
> > How do multiple interfaces (vifs) work in rtw88 if the port is always
> > zero? Is some kind of tracking of the used ports missing (similar to
> > how we track the used station IDs - also called mac_id - in
> > rtw_dev->mac_id_map)?
>
> The port should be allocated dynamically if we support two or more vifs.
> We have internal tree that is going to support p2p by second vif.
I see, thanks for clarifying this!


Best regards,
Martin

Hi,

On Sun, 2022-01-30 at 22:40 +0100, Martin Blumenstingl wrote:
> 
> On Fri, Jan 28, 2022 at 1:51 AM Pkshih <pkshih@realtek.com> wrote:
> [...]
> > > > To avoid this, we can add a flag to struct rtw_vif, and set this flag
> > > > when ::remove_interface. Then, only collect vif without this flag into list
> > > > when we use iterate_actiom().
> > > > 
> > > > As well as ieee80211_sta can do similar fix.
> > > > 
> > 
> > I would prefer my method that adds a 'bool disabled' flag to struct rtw_vif/rtw_sta
> > and set it when ::remove_interface/::sta_remove. Then rtw_iterate_stas() can
> > check this flag to decide whether does thing or not.
> That would indeed be a very straight forward approach and easy to read.
> In net/mac80211/iface.c there's some cases where after
> drv_remove_interface() (which internally calls our .remove_interface
> op) will kfree the vif (sdata). Doesn't that then result in a
> use-after-free if we rely on a boolean within rtw_vif?

The rtw_vif is drv_priv of ieee80211_vif, and they will be freed at
the same time. We must set 'bool disabled' after holding rtwdev->mutex
lock, and check this flag in iterator of ieee80211_iterate_active_interfaces_atomic()
to contruct a list of vif.

That means we never access this flag out of rtwdev->mutx or iterator.
Does it make sense?

--
Ping-Ke