2020-02-19 09:13:01

by Amol Grover

Subject: [PATCH] cfg80211: Pass lockdep expression to RCU lists

rdev->sched_scan_req_list maybe traversed using list_for_each_entry_rcu
outside an RCU read-side critical section but under the protection
of rtnl_mutex.

Hence, add corresponding lockdep expression to silence false-positive
warnings, and harden RCU lists.

Signed-off-by: Amol Grover <[email protected]>
---
net/wireless/scan.c | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/net/wireless/scan.c b/net/wireless/scan.c
index aef240fdf8df..7f1af8f347b1 100644
--- a/net/wireless/scan.c
+++ b/net/wireless/scan.c
@@ -556,9 +556,8 @@ cfg80211_find_sched_scan_req(struct cfg80211_registered_device *rdev, u64 reqid)
{
struct cfg80211_sched_scan_request *pos;

- WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
-
- list_for_each_entry_rcu(pos, &rdev->sched_scan_req_list, list) {
+ list_for_each_entry_rcu(pos, &rdev->sched_scan_req_list, list,
+ lockdep_rtnl_is_held()) {
if (pos->reqid == reqid)
return pos;
}
--
2.24.1

2020-02-19 09:15:28

by Johannes Berg

Subject: Re: [PATCH] cfg80211: Pass lockdep expression to RCU lists

On Wed, 2020-02-19 at 14:41 +0530, Amol Grover wrote:
>
> - WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
> -
> - list_for_each_entry_rcu(pos, &rdev->sched_scan_req_list, list) {
> + list_for_each_entry_rcu(pos, &rdev->sched_scan_req_list, list,
> + lockdep_rtnl_is_held()) {

Huh, I didn't even know you _could_ do that :)

johannes

2020-02-19 12:47:34

by Paul E. McKenney

Subject: Re: [PATCH] cfg80211: Pass lockdep expression to RCU lists

On Wed, Feb 19, 2020 at 10:13:36AM +0100, Johannes Berg wrote:
> On Wed, 2020-02-19 at 14:41 +0530, Amol Grover wrote:
> >
> > - WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
> > -
> > - list_for_each_entry_rcu(pos, &rdev->sched_scan_req_list, list) {
> > + list_for_each_entry_rcu(pos, &rdev->sched_scan_req_list, list,
> > + lockdep_rtnl_is_held()) {
>
> Huh, I didn't even know you _could_ do that :)

It is a fairly recent addition, courtesy of Joel Fernandes. ;-)

Thanx, Paul

2020-02-19 21:28:06

by Arend van Spriel

Subject: Re: [PATCH] cfg80211: Pass lockdep expression to RCU lists

On 2/19/2020 10:13 AM, Johannes Berg wrote:
> On Wed, 2020-02-19 at 14:41 +0530, Amol Grover wrote:
>>
>> - WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
>> -
>> - list_for_each_entry_rcu(pos, &rdev->sched_scan_req_list, list) {
>> + list_for_each_entry_rcu(pos, &rdev->sched_scan_req_list, list,
>> + lockdep_rtnl_is_held()) {
>
> Huh, I didn't even know you _could_ do that :)

Me neither ;-). Above you are removing the WARN_ON_ONCE() entirely.
Would it not be good to keep the WARN_ON_ONCE() with only the
!rcu_read_lock_held() check.

Regards,
Arend

2020-02-19 21:30:18

by Johannes Berg

Subject: Re: [PATCH] cfg80211: Pass lockdep expression to RCU lists

On Wed, 2020-02-19 at 22:27 +0100, Arend Van Spriel wrote:
> On 2/19/2020 10:13 AM, Johannes Berg wrote:
> > On Wed, 2020-02-19 at 14:41 +0530, Amol Grover wrote:
> > >
> > > - WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
> > > -
> > > - list_for_each_entry_rcu(pos, &rdev->sched_scan_req_list, list) {
> > > + list_for_each_entry_rcu(pos, &rdev->sched_scan_req_list, list,
> > > + lockdep_rtnl_is_held()) {
> >
> > Huh, I didn't even know you _could_ do that :)
>
> Me neither ;-). Above you are removing the WARN_ON_ONCE() entirely.
> Would it not be good to keep the WARN_ON_ONCE() with only the
> !rcu_read_lock_held() check.

Not needed, the macro expansion will already contain
rcu_read_lock_any_held() just like in all the other cases where you pass
a lockdep condition to RCU helpers.

johannes

2020-02-19 21:35:08

by Arend van Spriel

Subject: Re: [PATCH] cfg80211: Pass lockdep expression to RCU lists

On 2/19/2020 10:29 PM, Johannes Berg wrote:
> On Wed, 2020-02-19 at 22:27 +0100, Arend Van Spriel wrote:
>> On 2/19/2020 10:13 AM, Johannes Berg wrote:
>>> On Wed, 2020-02-19 at 14:41 +0530, Amol Grover wrote:
>>>>
>>>> - WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
>>>> -
>>>> - list_for_each_entry_rcu(pos, &rdev->sched_scan_req_list, list) {
>>>> + list_for_each_entry_rcu(pos, &rdev->sched_scan_req_list, list,
>>>> + lockdep_rtnl_is_held()) {
>>>
>>> Huh, I didn't even know you _could_ do that :)
>>
>> Me neither ;-). Above you are removing the WARN_ON_ONCE() entirely.
>> Would it not be good to keep the WARN_ON_ONCE() with only the
>> !rcu_read_lock_held() check.
>
> Not needed, the macro expansion will already contain
> rcu_read_lock_any_held() just like in all the other cases where you pass
> a lockdep condition to RCU helpers.

Ah, yes. I see it in __list_check_rcu().

Thanks,
Arend