Re: [PATCH wireless] wifi: mac8021: fix possible oob access in ieee80211_get_rate_duration

["Toke Høiland-Jørgensen" <toke@toke.dk>]

Lorenzo Bianconi <lorenzo@kernel.org> writes:

>> Lorenzo Bianconi <lorenzo@kernel.org> writes:
>> 
>> >> Lorenzo Bianconi <lorenzo@kernel.org> writes:
>> >> 
>> >> > Fix possible out-of-bound access in ieee80211_get_rate_duration routine
>> >> > as reported by the following UBSAN report:
>> >> >
>> >> > UBSAN: array-index-out-of-bounds in net/mac80211/airtime.c:455:47
>> >> > index 15 is out of range for type 'u16 [12]'
>> >> > CPU: 2 PID: 217 Comm: kworker/u32:10 Not tainted 6.1.0-060100rc3-generic
>> >> > Hardware name: Acer Aspire TC-281/Aspire TC-281, BIOS R01-A2 07/18/2017
>> >> > Workqueue: mt76 mt76u_tx_status_data [mt76_usb]
>> >> > Call Trace:
>> >> >  <TASK>
>> >> >  show_stack+0x4e/0x61
>> >> >  dump_stack_lvl+0x4a/0x6f
>> >> >  dump_stack+0x10/0x18
>> >> >  ubsan_epilogue+0x9/0x43
>> >> >  __ubsan_handle_out_of_bounds.cold+0x42/0x47
>> >> > ieee80211_get_rate_duration.constprop.0+0x22f/0x2a0 [mac80211]
>> >> >  ? ieee80211_tx_status_ext+0x32e/0x640 [mac80211]
>> >> >  ieee80211_calc_rx_airtime+0xda/0x120 [mac80211]
>> >> >  ieee80211_calc_tx_airtime+0xb4/0x100 [mac80211]
>> >> >  mt76x02_send_tx_status+0x266/0x480 [mt76x02_lib]
>> >> >  mt76x02_tx_status_data+0x52/0x80 [mt76x02_lib]
>> >> >  mt76u_tx_status_data+0x67/0xd0 [mt76_usb]
>> >> >  process_one_work+0x225/0x400
>> >> >  worker_thread+0x50/0x3e0
>> >> >  ? process_one_work+0x400/0x400
>> >> >  kthread+0xe9/0x110
>> >> >  ? kthread_complete_and_exit+0x20/0x20
>> >> >  ret_from_fork+0x22/0x30
>> >> >
>> >> > Reported-by: bjlockie@lockie.ca
>> >> > Fixes: db3e1c40cf2f ("mac80211: Import airtime calculation code from mt76")
>> >> > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
>> >> > ---
>> >> >  net/mac80211/airtime.c | 3 +++
>> >> >  1 file changed, 3 insertions(+)
>> >> >
>> >> > diff --git a/net/mac80211/airtime.c b/net/mac80211/airtime.c
>> >> > index 2e66598fac79..4ed05988131d 100644
>> >> > --- a/net/mac80211/airtime.c
>> >> > +++ b/net/mac80211/airtime.c
>> >> > @@ -452,6 +452,9 @@ static u32 ieee80211_get_rate_duration(struct ieee80211_hw *hw,
>> >> >  			 (status->encoding == RX_ENC_HE && streams > 8)))
>> >> >  		return 0;
>> >> >  
>> >> > +	if (WARN_ON_ONCE(idx >= MCS_GROUP_RATES))
>> >> > +		return 0;
>> >> > +
>> >> 
>> >> So presumably this is something that can actually happen in real usage,
>> >> so should we really warn? Or was the driver also fixed to not trigger
>> >> this?
>> >
>> > looking at the mt76x02 support, MT_RATE_INDEX_VHT_IDX is GENMASK(3, 0) so the
>> > hw can report rate_idx up to 15. Do you prefer to drop WARN_ON_ONCE()? I would
>> > prefer to keep it since it informs us something nasty occurred (and at the end
>> > it just runs ones), but I can live even w/o it :)
>> 
>> Well, what I mean is that the purpose of WARN_ON is, as you say, to
>> catch if "something nasty occurred", so we can fix it. But if we already
>> know that something nasty does, indeed, occur, shouldn't we just fix the
>> cause instead of putting in a warn so that we'll get a spat the next
>> time it happens? :)
>
> I think in this case the hw just reports a wrong value, so we can limit the
> value there too, anyway I would not assume each driver limits the rate_idx
> value (as we already do for stream :))

Right, I'm not disputing we should add the check itself :)

I guess we could phrase it like this: Do we expect drivers to be fixed
to clamp the index to < MCS_GROUP_RATES? If so, keep the warn; if we
don't expect drivers to be fixed to handle this, drop the warn and just
handle it gracefully...

-Toke