[B.A.T.M.A.N.] [PATCH maint] batman-adv: Initialize memory for station_info

List overview All Threads
Download

newer

older

[B.A.T.M.A.N.] [PATCH maint v2]...

[B.A.T.M.A.N.] [PATCH maint 1/2]...

Sven Eckelmann

5 Jun 2018 5 Jun '18

6:31 p.m.

cfg80211_get_station is not initializing the memory given as parameter sinfo. The caller has to handle it. Otherwise the filled parameter may be set incorrectly and thus uninitialized memory is used to identify the throughput to an neighbor.

Fixes: 5c3245172c01 ("batman-adv: ELP - compute the metric based on the estimated throughput") Reported-by: Thomas Lauer holminateur@gmail.com Reported-by: Marcel Schmidt ff.z-casparistrasse@mailbox.org Signed-off-by: Sven Eckelmann sven@narfation.org ---

Cc: Thomas Lauer holminateur@gmail.com Cc: Marcel Schmidt ff.z-casparistrasse@mailbox.org

net/batman-adv/bat_v_elp.c | 1 + 1 file changed, 1 insertion(+)

diff --git a/net/batman-adv/bat_v_elp.c b/net/batman-adv/bat_v_elp.c index 71c20c1d..5f931475 100644 --- a/net/batman-adv/bat_v_elp.c +++ b/net/batman-adv/bat_v_elp.c @@ -102,6 +102,7 @@ static u32 batadv_v_elp_get_throughput(struct batadv_hardif_neigh_node *neigh) if (!real_netdev) goto default_throughput;

+ memset(&sinfo, 0, sizeof(sinfo)); ret = cfg80211_get_station(real_netdev, neigh->addr, &sinfo);

dev_put(real_netdev);

-- 2.11.0

Show replies by date

Sven Eckelmann

5 Jun 5 Jun

10:01 p.m.

Hi Antonio,

On Dienstag, 5. Juni 2018 20:31:31 CEST Sven Eckelmann wrote:

...

cfg80211_get_station is not initializing the memory given as parameter sinfo. The caller has to handle it. Otherwise the filled parameter may be set incorrectly and thus uninitialized memory is used to identify the throughput to an neighbor.

Hm, have to rewrite the commit message to something better. Maybe to something like:

batadv_v_elp_get_throughput is calling cfg80211_get_station with a pointer (sinfo) to some uninitialized memory on the stack. But most of the implementations behind cfg80211_get_station will not initialize sinfo to zero before manipulating it. For example, the member &struct station_info.filled is often only modified by using a read (of possibly uninitialized/random memory), an OR operation and then a write of the new value back to the original memory address. A caller without a preinitialized &struct station_info.filled can then no longer decide which parts of sinfo were filled in by cfg80211_get_station.

The caller of cfg80211_get_station must therefore take care that sinfo (or at least sinfo.filled) is initialized to zero. Otherwise, the caller may tries to read information which was not filled in and is therefore also uninitialized. In batadv_v_elp_get_throughput's case, an invalid "random" expected throughput may be saved for this neighbor and thus the B.A.T.M.A.N V algorithm may switch to non-optimal neighbors for certain destinations.

...

Fixes: 5c3245172c01 ("batman-adv: ELP - compute the metric based on the estimated throughput") Reported-by: Thomas Lauer holminateur@gmail.com Reported-by: Marcel Schmidt ff.z-casparistrasse@mailbox.org Signed-off-by: Sven Eckelmann sven@narfation.org

Cc: Thomas Lauer holminateur@gmail.com Cc: Marcel Schmidt ff.z-casparistrasse@mailbox.org

Here some more information why I thought that this could be the reason (still has to be tested by the reporters):

* sinfo is usually allocated with alloc helper that initializes everything to 0 [1,2] * filled is set to 0 by functions which have sinfo on stack and call the (internal) functionality [3,4] * most implementations for cfg80211_get_station (&struct cfg80211_ops->get_station) [5] are *not* setting filled manually to a predefined value - default implementation [6] and its main helper [7] doesn't do that - ath6kl [8] doesn't do that - (ath10k [9] doesn't do that for statistics) - wil6210 [10] does it - brcmfmac does it for everything [11] except ibss [12] - (iwlwifi doesn't do that for the statistics [13]) - libertas doesn't do that [14] - mwifiex does it [15] - quantenna doesn't do it [16,17] - (wlcore doesn't do it [18] for statistics) - rndis_wlan is doing it [19] - rtl8723bs does it [20] - (rtlwifi does it for statistics [21]) - wilc1000 doesn't do it [22] - prism2 does it [23] * nl80211 interface also does the memset [24] * wext compat layer does it everywhere(?) [25]

But please check this again because a lot of the "research" was done in a restaurant.

Kind regards, Sven

Antonio Quartulli

6 Jun 6 Jun

4:44 a.m.

Hi,

On 06/06/18 06:01, Sven Eckelmann wrote:

...

Hi Antonio,

On Dienstag, 5. Juni 2018 20:31:31 CEST Sven Eckelmann wrote:

...
cfg80211_get_station is not initializing the memory given as parameter sinfo. The caller has to handle it. Otherwise the filled parameter may be set incorrectly and thus uninitialized memory is used to identify the throughput to an neighbor.

Hm, have to rewrite the commit message to something better. Maybe to something like:

batadv_v_elp_get_throughput is calling cfg80211_get_station with a pointer (sinfo) to some uninitialized memory on the stack. But most of the implementations behind cfg80211_get_station will not initialize sinfo to zero before manipulating it. For example, the member &struct station_info.filled is often only modified by using a read (of possibly uninitialized/random memory), an OR operation and then a write of the new value back to the original memory address. A caller without a preinitialized &struct station_info.filled can then no longer decide which parts of sinfo were filled in by cfg80211_get_station.

The caller of cfg80211_get_station must therefore take care that sinfo (or at least sinfo.filled) is initialized to zero. Otherwise, the caller may tries to read information which was not filled in and is therefore also uninitialized. In batadv_v_elp_get_throughput's case, an invalid "random" expected throughput may be saved for this neighbor and thus the B.A.T.M.A.N V algorithm may switch to non-optimal neighbors for certain destinations.

...
Fixes: 5c3245172c01 ("batman-adv: ELP - compute the metric based on the estimated throughput") Reported-by: Thomas Lauer holminateur@gmail.com Reported-by: Marcel Schmidt ff.z-casparistrasse@mailbox.org Signed-off-by: Sven Eckelmann sven@narfation.org

This looks fairly reasonable and at a first glance I can confirm that even ieee80211_get_station() does not clear it up before filling it.

However, what do you think about zero'ing the sinfo object directly in cfg80211_get_station()? I think it is safe to assume that no user should store anything in this object before passing it to get_station().

At the same time future users of this API won't be able to hit the same issue we just did.

(At the moment we are the only user of this function in the kernel, therefore changing its behaviour now might still be an option)

Cheers,

-- Antonio Quartulli

Sven Eckelmann

5:25 a.m.

On Mittwoch, 6. Juni 2018 06:44:32 CEST Antonio Quartulli wrote: [...]

...

This looks fairly reasonable and at a first glance I can confirm that even ieee80211_get_station() does not clear it up before filling it.

However, what do you think about zero'ing the sinfo object directly in cfg80211_get_station()? I think it is safe to assume that no user should store anything in this object before passing it to get_station().

I understand your point. But there is the potential problem that such a change requires more discussion with the Johannes/other wireless guys - while the early B.A.T.M.A.N. V adopters still suffer under this problem. I will ask him in IRC first about his opinion but send a second version of the patch to the batman-adv mailing list. We can later decide which route we take.

Kind regards, Sven

2361

Age (days ago)

2362

Last active (days ago)

b.a.t.m.a.n@lists.open-mesh.org

3 comments

2 participants

tags (0)

participants (2)

Antonio Quartulli
Sven Eckelmann