diff mbox series

net: mac80211: cfg: enforce sanity checks for key_index in ieee80211_del_key()

Message ID 20201201095639.63936-1-anant.thazhemadam@gmail.com
State New
Headers show
Series net: mac80211: cfg: enforce sanity checks for key_index in ieee80211_del_key() | expand

Commit Message

Anant Thazhemadam Dec. 1, 2020, 9:56 a.m. UTC
Currently, it is assumed that key_idx values that are passed to
ieee80211_del_key() are all valid indexes as is, and no sanity checks
are performed for it.
However, syzbot was able to trigger an array-index-out-of-bounds bug
by passing a key_idx value of 5, when the maximum permissible index
value is (NUM_DEFAULT_KEYS - 1).
Enforcing sanity checks helps in preventing this bug, or a similar
instance in the context of ieee80211_del_key() from occurring.

Reported-by: syzbot+49d4cab497c2142ee170@syzkaller.appspotmail.com
Tested-by: syzbot+49d4cab497c2142ee170@syzkaller.appspotmail.com
Signed-off-by: Anant Thazhemadam <anant.thazhemadam@gmail.com>
---
 net/mac80211/cfg.c | 24 +++++++++++++++++++++---
 1 file changed, 21 insertions(+), 3 deletions(-)

Comments

Johannes Berg Dec. 1, 2020, 10 a.m. UTC | #1
On Tue, 2020-12-01 at 15:26 +0530, Anant Thazhemadam wrote:
> Currently, it is assumed that key_idx values that are passed to
> ieee80211_del_key() are all valid indexes as is, and no sanity checks
> are performed for it.
> However, syzbot was able to trigger an array-index-out-of-bounds bug
> by passing a key_idx value of 5, when the maximum permissible index
> value is (NUM_DEFAULT_KEYS - 1).
> Enforcing sanity checks helps in preventing this bug, or a similar
> instance in the context of ieee80211_del_key() from occurring.

I think we should do this more generally in cfg80211, like in
nl80211_new_key() we do it via cfg80211_validate_key_settings().

I suppose we cannot use the same function, but still, would be good to
address this generally in nl80211 for all drivers.

johannes
Anant Thazhemadam Dec. 1, 2020, 11:56 a.m. UTC | #2
On 01/12/20 3:30 pm, Johannes Berg wrote:
> On Tue, 2020-12-01 at 15:26 +0530, Anant Thazhemadam wrote:
>> Currently, it is assumed that key_idx values that are passed to
>> ieee80211_del_key() are all valid indexes as is, and no sanity checks
>> are performed for it.
>> However, syzbot was able to trigger an array-index-out-of-bounds bug
>> by passing a key_idx value of 5, when the maximum permissible index
>> value is (NUM_DEFAULT_KEYS - 1).
>> Enforcing sanity checks helps in preventing this bug, or a similar
>> instance in the context of ieee80211_del_key() from occurring.
> I think we should do this more generally in cfg80211, like in
> nl80211_new_key() we do it via cfg80211_validate_key_settings().
>
> I suppose we cannot use the same function, but still, would be good to
> address this generally in nl80211 for all drivers.

Hello,

This gave me the idea of trying to use cfg80211_validate_key_settings()
directly in ieee80211_del_key(). I did try that out, tested it, and this bug
doesn't seem to be getting triggered anymore.
If this is okay, then I can send in a v2 soon. :)

If there is any reason that I'm missing as to why cfg80211_validate_key_settings()
cannot be used in this context, please let me know.

Thanks,
Anant
Johannes Berg Dec. 1, 2020, 12:06 p.m. UTC | #3
On Tue, 2020-12-01 at 17:26 +0530, Anant Thazhemadam wrote:
> On 01/12/20 3:30 pm, Johannes Berg wrote:
> > On Tue, 2020-12-01 at 15:26 +0530, Anant Thazhemadam wrote:
> > > Currently, it is assumed that key_idx values that are passed to
> > > ieee80211_del_key() are all valid indexes as is, and no sanity checks
> > > are performed for it.
> > > However, syzbot was able to trigger an array-index-out-of-bounds bug
> > > by passing a key_idx value of 5, when the maximum permissible index
> > > value is (NUM_DEFAULT_KEYS - 1).
> > > Enforcing sanity checks helps in preventing this bug, or a similar
> > > instance in the context of ieee80211_del_key() from occurring.
> > I think we should do this more generally in cfg80211, like in
> > nl80211_new_key() we do it via cfg80211_validate_key_settings().
> > 
> > I suppose we cannot use the same function, but still, would be good to
> > address this generally in nl80211 for all drivers.
> 
> Hello,
> 
> This gave me the idea of trying to use cfg80211_validate_key_settings()
> directly in ieee80211_del_key(). I did try that out, tested it, and this bug
> doesn't seem to be getting triggered anymore.
> If this is okay, then I can send in a v2 soon. :)
> 
> If there is any reason that I'm missing as to why cfg80211_validate_key_settings()
> cannot be used in this context, please let me know.

If it works then I guess that's OK. I thought we didn't have all the
right information, e.g. whether a key is pairwise or not?

johannes
Anant Thazhemadam Dec. 1, 2020, 12:45 p.m. UTC | #4
On 01/12/20 5:36 pm, Johannes Berg wrote:
> On Tue, 2020-12-01 at 17:26 +0530, Anant Thazhemadam wrote:
>> On 01/12/20 3:30 pm, Johannes Berg wrote:
>>> On Tue, 2020-12-01 at 15:26 +0530, Anant Thazhemadam wrote:
>>>> Currently, it is assumed that key_idx values that are passed to
>>>> ieee80211_del_key() are all valid indexes as is, and no sanity checks
>>>> are performed for it.
>>>> However, syzbot was able to trigger an array-index-out-of-bounds bug
>>>> by passing a key_idx value of 5, when the maximum permissible index
>>>> value is (NUM_DEFAULT_KEYS - 1).
>>>> Enforcing sanity checks helps in preventing this bug, or a similar
>>>> instance in the context of ieee80211_del_key() from occurring.
>>> I think we should do this more generally in cfg80211, like in
>>> nl80211_new_key() we do it via cfg80211_validate_key_settings().
>>>
>>> I suppose we cannot use the same function, but still, would be good to
>>> address this generally in nl80211 for all drivers.
>> Hello,
>>
>> This gave me the idea of trying to use cfg80211_validate_key_settings()
>> directly in ieee80211_del_key(). I did try that out, tested it, and this bug
>> doesn't seem to be getting triggered anymore.
>> If this is okay, then I can send in a v2 soon. :)
>>
>> If there is any reason that I'm missing as to why cfg80211_validate_key_settings()
>> cannot be used in this context, please let me know.
> If it works then I guess that's OK. I thought we didn't have all the
> right information, e.g. whether a key is pairwise or not?
>
> johannes
>
Well,
cfg80211_supported_cipher_suite(&rdev->wiphy, params->cipher) returned
false, and thus it worked for the syzbot reproducer.
Would it be a safer idea to enforce the conditions that I initially put (in
ieee80211_del_key()) directly in cfg80211_validate_key_settings() itself - by
updating max_key_index, and checking accordingly?

Thanks,
Anant
Johannes Berg Dec. 1, 2020, 12:56 p.m. UTC | #5
On Tue, 2020-12-01 at 18:15 +0530, Anant Thazhemadam wrote:
> 
> cfg80211_supported_cipher_suite(&rdev->wiphy, params->cipher) returned
> false, and thus it worked for the syzbot reproducer.
> Would it be a safer idea to enforce the conditions that I initially put (in
> ieee80211_del_key()) directly in cfg80211_validate_key_settings() itself - by
> updating max_key_index, and checking accordingly?

Yes, I think so. But similarly to cfg80211_validate_key_settings() it
should look at the device capabilities (beacon protection, etc.)

Thanks!
johannes
diff mbox series

Patch

diff --git a/net/mac80211/cfg.c b/net/mac80211/cfg.c
index 7276e66ae435..d349e33134e6 100644
--- a/net/mac80211/cfg.c
+++ b/net/mac80211/cfg.c
@@ -516,12 +516,30 @@  static int ieee80211_del_key(struct wiphy *wiphy, struct net_device *dev,
 		if (!sta)
 			goto out_unlock;
 
-		if (pairwise)
+		if (pairwise) {
+			if (key_idx >= NUM_DEFAULT_KEYS) {
+				ret = -EINVAL;
+				goto out_unlock;
+			}
 			key = key_mtx_dereference(local, sta->ptk[key_idx]);
-		else
+		} else {
+			if (key_idx >= (NUM_DEFAULT_KEYS +
+					NUM_DEFAULT_MGMT_KEYS +
+					NUM_DEFAULT_BEACON_KEYS)) {
+				ret = -EINVAL;
+				goto out_unlock;
+			}
 			key = key_mtx_dereference(local, sta->gtk[key_idx]);
-	} else
+		}
+	} else {
+		if (key_idx >= (NUM_DEFAULT_KEYS +
+				NUM_DEFAULT_MGMT_KEYS +
+				NUM_DEFAULT_BEACON_KEYS)) {
+			ret = -EINVAL;
+			goto out_unlock;
+		}
 		key = key_mtx_dereference(local, sdata->keys[key_idx]);
+	}
 
 	if (!key) {
 		ret = -ENOENT;