Our detector has identified another case of an incomplete policy.
Specifically, the commit df78a0c0b67d ("nl80211: S1G band and channel
definitions") introduced the NL80211_BAND_S1GHZ attribute to
nl80211_band, but it neglected to update the
nl80211_match_band_rssi_policy accordingly.
Similar commits that add new band types, such as the initial
commit 1e1b11b6a111 ("nl80211/cfg80211: Specify band specific min RSSI
thresholds with sched scan"), the commit e548a1c36b11 ("cfg80211: add 6GHz
in code handling array with NUM_NL80211_BANDS entries"), and the
commit 63fa04266629 ("nl80211: Add LC placeholder band definition to
nl80211_band"), all require updates to the policy.
Failure to do so could result in accessing an attribute of unexpected
length in the function nl80211_parse_sched_scan_per_band_rssi.
To resolve this issue, this commit adds the policy for the
NL80211_BAND_S1GHZ attribute.
Fixes: df78a0c0b67d ("nl80211: S1G band and channel definitions")
Signed-off-by: Lin Ma <[email protected]>
---
V1 -> V2: change net-next to wireless as suggested
net/wireless/nl80211.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/net/wireless/nl80211.c b/net/wireless/nl80211.c
index 60877b532993..980300621a60 100644
--- a/net/wireless/nl80211.c
+++ b/net/wireless/nl80211.c
@@ -911,6 +911,7 @@ nl80211_match_band_rssi_policy[NUM_NL80211_BANDS] = {
[NL80211_BAND_5GHZ] = { .type = NLA_S32 },
[NL80211_BAND_6GHZ] = { .type = NLA_S32 },
[NL80211_BAND_60GHZ] = { .type = NLA_S32 },
+ [NL80211_BAND_S1GHZ] = { .type = NLA_S32 },
[NL80211_BAND_LC] = { .type = NLA_S32 },
};
--
2.34.1
On Fri, 2024-01-19 at 23:12 +0800, Lin Ma wrote:
>
> V1 -> V2: change net-next to wireless as suggested
For the record, I don't think anyone is using scheduled scan with S1G or
even _could_ be doing that (is there even a driver other than hwsim?),
so I don't think I'll apply this for wireless. No need to resend (again)
though :)
johannes
On 1/19/2024 7:12 AM, Lin Ma wrote:
> Our detector has identified another case of an incomplete policy.
> Specifically, the commit df78a0c0b67d ("nl80211: S1G band and channel
> definitions") introduced the NL80211_BAND_S1GHZ attribute to
> nl80211_band, but it neglected to update the
> nl80211_match_band_rssi_policy accordingly.
>
> Similar commits that add new band types, such as the initial
> commit 1e1b11b6a111 ("nl80211/cfg80211: Specify band specific min RSSI
> thresholds with sched scan"), the commit e548a1c36b11 ("cfg80211: add 6GHz
> in code handling array with NUM_NL80211_BANDS entries"), and the
> commit 63fa04266629 ("nl80211: Add LC placeholder band definition to
> nl80211_band"), all require updates to the policy.
> Failure to do so could result in accessing an attribute of unexpected
> length in the function nl80211_parse_sched_scan_per_band_rssi.
>
> To resolve this issue, this commit adds the policy for the
> NL80211_BAND_S1GHZ attribute.
>
> Fixes: df78a0c0b67d ("nl80211: S1G band and channel definitions")
> Signed-off-by: Lin Ma <[email protected]>
> ---
> V1 -> V2: change net-next to wireless as suggested
>
> net/wireless/nl80211.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/net/wireless/nl80211.c b/net/wireless/nl80211.c
> index 60877b532993..980300621a60 100644
> --- a/net/wireless/nl80211.c
> +++ b/net/wireless/nl80211.c
> @@ -911,6 +911,7 @@ nl80211_match_band_rssi_policy[NUM_NL80211_BANDS] = {
> [NL80211_BAND_5GHZ] = { .type = NLA_S32 },
> [NL80211_BAND_6GHZ] = { .type = NLA_S32 },
> [NL80211_BAND_60GHZ] = { .type = NLA_S32 },
> + [NL80211_BAND_S1GHZ] = { .type = NLA_S32 },
> [NL80211_BAND_LC] = { .type = NLA_S32 },
> };
>
something is really suspicious since the NL80211_BAND_* enums are
*value* enums, not attribute ID enums, and hence they should never be
used in an nla_policy.
what is actually using these as attribute IDs, noting that
NL80211_BAND_2GHZ == 0 and hence cannot be used as an attribute ID
seems the logic that introduced this policy needs to be revisited.
On Fri, 2024-01-19 at 15:47 -0800, Jeff Johnson wrote:
> > --- a/net/wireless/nl80211.c
> > +++ b/net/wireless/nl80211.c
> > @@ -911,6 +911,7 @@ nl80211_match_band_rssi_policy[NUM_NL80211_BANDS] = {
> > [NL80211_BAND_5GHZ] = { .type = NLA_S32 },
> > [NL80211_BAND_6GHZ] = { .type = NLA_S32 },
> > [NL80211_BAND_60GHZ] = { .type = NLA_S32 },
> > + [NL80211_BAND_S1GHZ] = { .type = NLA_S32 },
> > [NL80211_BAND_LC] = { .type = NLA_S32 },
> > };
> >
> something is really suspicious since the NL80211_BAND_* enums are
> *value* enums, not attribute ID enums, and hence they should never be
> used in an nla_policy.
Yeah, that's what it looks like first, but then they do get used
anyway...
> what is actually using these as attribute IDs, noting that
> NL80211_BAND_2GHZ == 0 and hence cannot be used as an attribute ID
Ohh. Good catch!
> seems the logic that introduced this policy needs to be revisited.
>
Let's just remove it?
commit 1e1b11b6a1111cd9e8af1fd6ccda270a9fa3eacf
Author: vamsi krishna <[email protected]>
Date: Fri Feb 1 18:34:51 2019 +0530
nl80211/cfg80211: Specify band specific min RSSI thresholds with sched scan
As far as I can tell nothing is using that in the first place ...
Certainly not in the kernel, nor wpa_s, nor anything else I could find
really ...
We can't completely revert it since we need the attribute number to stay
allocated, but that's all we cannot remove.
johannes
On 1/22/2024 10:33 AM, Jeff Johnson wrote:
> On 1/20/2024 12:27 PM, Johannes Berg wrote:
>> On Fri, 2024-01-19 at 15:47 -0800, Jeff Johnson wrote:
>>>> --- a/net/wireless/nl80211.c
>>>> +++ b/net/wireless/nl80211.c
>>>> @@ -911,6 +911,7 @@ nl80211_match_band_rssi_policy[NUM_NL80211_BANDS] = {
>>>> [NL80211_BAND_5GHZ] = { .type = NLA_S32 },
>>>> [NL80211_BAND_6GHZ] = { .type = NLA_S32 },
>>>> [NL80211_BAND_60GHZ] = { .type = NLA_S32 },
>>>> + [NL80211_BAND_S1GHZ] = { .type = NLA_S32 },
>>>> [NL80211_BAND_LC] = { .type = NLA_S32 },
>>>> };
>>>>
>>> something is really suspicious since the NL80211_BAND_* enums are
>>> *value* enums, not attribute ID enums, and hence they should never be
>>> used in an nla_policy.
>>
>> Yeah, that's what it looks like first, but then they do get used
>> anyway...
>>
>>> what is actually using these as attribute IDs, noting that
>>> NL80211_BAND_2GHZ == 0 and hence cannot be used as an attribute ID
>>
>> Ohh. Good catch!
>>
>>> seems the logic that introduced this policy needs to be revisited.
>>>
>>
>> Let's just remove it?
>>
>> commit 1e1b11b6a1111cd9e8af1fd6ccda270a9fa3eacf
>> Author: vamsi krishna <[email protected]>
>> Date: Fri Feb 1 18:34:51 2019 +0530
>>
>> nl80211/cfg80211: Specify band specific min RSSI thresholds with sched scan
>>
>>
>> As far as I can tell nothing is using that in the first place ...
>> Certainly not in the kernel, nor wpa_s, nor anything else I could find
>> really ...
>>
>> We can't completely revert it since we need the attribute number to stay
>> allocated, but that's all we cannot remove.
>
> I'm investigating this and will report back.
OK, I have investigated this and based upon the investigation this can
be removed (except for keeping the now obsolete uapi bits). This was
done in preparation for supporting a new Android interface in the
out-of-tree Android driver, but that interface was subsequently
withdrawn by Google.
Johannes, do you want to handle this? Or should I?
/jeff
On Thu, 2024-01-25 at 10:32 -0800, Jeff Johnson wrote:
>
> OK, I have investigated this and based upon the investigation this can
> be removed (except for keeping the now obsolete uapi bits). This was
> done in preparation for supporting a new Android interface in the
> out-of-tree Android driver, but that interface was subsequently
> withdrawn by Google.
>
> Johannes, do you want to handle this? Or should I?
Would be great if you could send a patch, thanks!
johannes