Re: [RFC] mac80211: handle -EXIST on sta_info_insert()

[Johannes Berg <johannes@sipsolutions.net>]

[-- Attachment #1: Type: text/plain, Size: 1472 bytes --]

On Mon, 2009-06-01 at 18:30 -0700, Luis R. Rodriguez wrote:
> On Mon, Jun 1, 2009 at 6:23 PM, Luis R. Rodriguez
> <lrodriguez@atheros.com> wrote:
> > There's a few places where we either did not rcu_read_lock()
> > prior to addition of a a new sta or we allocated it before
> > checking for its existance. In most places, like device open
> > and close we should have at least some guarantee the stas are
> > wiped but in other places this could not be the case.
> >
> > Lets protect against RCU in the missing places. The only
> > place I see is is in ieee80211_rx_bss_info(). Not sure
> > we are calling ieee80211_ibss_add_sta() twice there though.
> >
> > In our mac80211 cfg80211 callback for device addition we
> > also can simplify the code by first checking for the STA
> > before trying to add it and then checking for -EEXIST which
> > we were not doing. If that actualy would happen we could
> > end up potentially with a stale sta and the rate info was
> > never updated. It seems cleaner to check for the sta first.
> >
> > Lastly, we add a WARN_ON() on the STA mlme path upon call to
> > ieee80211_rx_mgmt_assoc_resp() for -EEXIST. This should not
> > happen, we could just return -EIO or simply ignore it.
> 
> Hm, actually on second thought what if we simply kdoc that you should
> check for the sta's existence first prior to addition. Then we can
> remove the pesky -EEXIST.

Umm, no? Neither approach is correct.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 801 bytes --]