[PATCH AUTOSEL for 3.18 030/101] net_sched: move tcf_lock down after gen_replace_estimator()

[Sasha Levin]

From: WANG Cong 

[ Upstream commit 74030603dfd9f76c0f279f19f1dd1ee3028fee7a ]

Laura reported a sleep-in-atomic kernel warning inside
tcf_act_police_init() which calls gen_replace_estimator() with
spinlock protection.

It is not necessary in this case, we already have RTNL lock here
so it is enough to protect concurrent writers. For the reader,
i.e. tcf_act_police(), it needs to make decision based on this
rate estimator, in the worst case we drop more/less packets than
necessary while changing the rate in parallel, it is still acceptable.

Reported-by: Laura Abbott 
Reported-by: Nick Huber 
Cc: Jamal Hadi Salim 
Signed-off-by: Cong Wang 
Acked-by: Jamal Hadi Salim 
Signed-off-by: David S. Miller 
Signed-off-by: Sasha Levin 
---
 net/sched/act_police.c | 8 +++-
 1 file changed, 3 insertions(+), 5 deletions(-)

diff --git a/net/sched/act_police.c b/net/sched/act_police.c
index 69791ca77a05..1a25d25b9a56 100644
--- a/net/sched/act_police.c
+++ b/net/sched/act_police.c
@@ -176,21 +176,21 @@ override:
}
}
 
-   spin_lock_bh(>tcf_lock);
if (est) {
err = gen_replace_estimator(>tcf_bstats, NULL,
>tcf_rate_est,
>tcf_lock, est);
if (err)
-   goto failure_unlock;
+   goto failure;
} else if (tb[TCA_POLICE_AVRATE] &&
   (ret == ACT_P_CREATED ||
!gen_estimator_active(>tcf_bstats,
  >tcf_rate_est))) {
err = -EINVAL;
-   goto failure_unlock;
+   goto failure;
}
 
+   spin_lock_bh(>tcf_lock);
/* No failure allowed after this point */
police->tcfp_mtu = parm->mtu;
if (police->tcfp_mtu == 0) {
@@ -242,8 +242,6 @@ override:
a->priv = police;
return ret;
 
-failure_unlock:
-   spin_unlock_bh(>tcf_lock);
 failure:
qdisc_put_rtab(P_tab);
qdisc_put_rtab(R_tab);
-- 
2.15.1

[PATCH AUTOSEL for 3.18 030/101] net_sched: move tcf_lock down after gen_replace_estimator()

[Sasha Levin]

From: WANG Cong 

[ Upstream commit 74030603dfd9f76c0f279f19f1dd1ee3028fee7a ]

Laura reported a sleep-in-atomic kernel warning inside
tcf_act_police_init() which calls gen_replace_estimator() with
spinlock protection.

It is not necessary in this case, we already have RTNL lock here
so it is enough to protect concurrent writers. For the reader,
i.e. tcf_act_police(), it needs to make decision based on this
rate estimator, in the worst case we drop more/less packets than
necessary while changing the rate in parallel, it is still acceptable.

Reported-by: Laura Abbott 
Reported-by: Nick Huber 
Cc: Jamal Hadi Salim 
Signed-off-by: Cong Wang 
Acked-by: Jamal Hadi Salim 
Signed-off-by: David S. Miller 
Signed-off-by: Sasha Levin 
---
 net/sched/act_police.c | 8 +++-
 1 file changed, 3 insertions(+), 5 deletions(-)

diff --git a/net/sched/act_police.c b/net/sched/act_police.c
index 69791ca77a05..1a25d25b9a56 100644
--- a/net/sched/act_police.c
+++ b/net/sched/act_police.c
@@ -176,21 +176,21 @@ override:
}
}
 
-   spin_lock_bh(>tcf_lock);
if (est) {
err = gen_replace_estimator(>tcf_bstats, NULL,
>tcf_rate_est,
>tcf_lock, est);
if (err)
-   goto failure_unlock;
+   goto failure;
} else if (tb[TCA_POLICE_AVRATE] &&
   (ret == ACT_P_CREATED ||
!gen_estimator_active(>tcf_bstats,
  >tcf_rate_est))) {
err = -EINVAL;
-   goto failure_unlock;
+   goto failure;
}
 
+   spin_lock_bh(>tcf_lock);
/* No failure allowed after this point */
police->tcfp_mtu = parm->mtu;
if (police->tcfp_mtu == 0) {
@@ -242,8 +242,6 @@ override:
a->priv = police;
return ret;
 
-failure_unlock:
-   spin_unlock_bh(>tcf_lock);
 failure:
qdisc_put_rtab(P_tab);
qdisc_put_rtab(R_tab);
-- 
2.15.1