ceph: cancel delayed work instead of flushing on mdsc teardown
PATCH v2 1/1 ceph: cancel delayed work instead of flushing on mdsc teardown
The first thing metric_delayed_work does is check mdsc->stopping, and then return immediately if it's set. That's good since we would have already torn down the metric structures at this point, otherwise, but there is no locking around mdsc->stopping.
It's possible that the ceph_metric_destroy call could race with the delayed_work, in which case we could end up with the delayed_work accessing destroyed percpu variables.
At this point in the mdsc teardown, the "stopping" flag has already been set, so there's no benefit to flushing the work. Move the work cancellation in ceph_metric_destroy ahead of the percpu variable destruction, and eliminate the flush_delayed_work call in ceph_mdsc_destroy.
--- fs/ceph/mds_client.c | 1 - fs/ceph/metric.c | 4 ++-- 2 files changed, 2 insertions(+), 3 deletions(-)
v2: just drop the flush call altogether and move the cancel before the
percpu variables are destroyed (per Xiubo's suggestion).
fs/ceph/mds_client.c +0 -1
@@ -4979,7 +4979,6 @@ void ceph_mdsc_destroy(struct ceph_fs_client *fsc)ceph_metric_destroy (&mdsc->metric); -flush_delayed_work (&mdsc->metric.delayed_work); fsc->mdsc =NULL ;kfree (mdsc);dout ("mdsc_destroy %p done\n" , mdsc);
fs/ceph/metric.c +2 -2
@@ -302,6 +302,8 @@ void ceph_metric_destroy(struct ceph_client_metric *m)if (!m)return ; +cancel_delayed_work_sync (&m->delayed_work); +percpu_counter_destroy (&m->total_inodes);percpu_counter_destroy (&m->opened_inodes);percpu_counter_destroy (&m->i_caps_mis);
@@ -309,8 +311,6 @@ void ceph_metric_destroy(struct ceph_client_metric *m)percpu_counter_destroy (&m->d_lease_mis);percpu_counter_destroy (&m->d_lease_hit); -cancel_delayed_work_sync (&m->delayed_work); -ceph_put_mds_session (m->session); }
Signature · 1 line
2.31.1
Quoted 52 lines · Jeff Layton
The first thing
metric_delayed_workdoes is checkmdsc->stopping, and then return immediately if it's set. That's good since we would have already torn down the metric structures at this point, otherwise, but there is no locking aroundmdsc->stopping.It's possible that the
ceph_metric_destroycall could race with thedelayed_work, in which case we could end up with thedelayed_workaccessing destroyed percpu variables.At this point in the mdsc teardown, the "stopping" flag has already been set, so there's no benefit to flushing the work. Move the work cancellation in
ceph_metric_destroyahead of the percpu variable destruction, and eliminate theflush_delayed_workcall inceph_mdsc_destroy.Cc Xiubo Li <[email protected]>Fixes 18f473b384a6 ("ceph: periodically send perf metrics to MDSes")Signed-off-by Jeff Layton <[email protected]>--- fs/ceph/mds_client.c | 1 - fs/ceph/metric.c | 4 ++-- 2 files changed, 2 insertions(+), 3 deletions(-)v2: just drop the flush call altogether and move the cancel before the
percpu variables are destroyed (per Xiubo's suggestion).fs/ceph/mds_client.c +0
-1@@ -4979,7 +4979,6 @@ void ceph_mdsc_destroy(struct ceph_fs_client *fsc)ceph_metric_destroy (&mdsc->metric); -flush_delayed_work (&mdsc->metric.delayed_work); fsc->mdsc =NULL ;kfree (mdsc);dout ("mdsc_destroy %p done\n" , mdsc);fs/ceph/metric.c +2
-0@@ -302,6 +302,8 @@ void ceph_metric_destroy(struct ceph_client_metric *m)if (!m)return ; +cancel_delayed_work_sync (&m->delayed_work); +percpu_counter_destroy (&m->total_inodes);percpu_counter_destroy (&m->opened_inodes);percpu_counter_destroy (&m->i_caps_mis);
fs/ceph/metric.c void ceph_metric_destroy(struct ceph_client_metric *m) · line 316
Quoted 8 lines
fs/ceph/metric.c +0
-2@@ -309,8 +311,6 @@ void ceph_metric_destroy(struct ceph_client_metric *m)percpu_counter_destroy (&m->d_lease_mis);percpu_counter_destroy (&m->d_lease_hit); -cancel_delayed_work_sync (&m->delayed_work); -ceph_put_mds_session (m->session); }