@@ -2064,6 +2064,23 @@ static int check_caps_flush(struct ceph_mds_client *mdsc,
return ret;
}
+static void dump_cap_flushes(struct ceph_mds_client *mdsc, u64 want_tid)
+{
+ int index = 0;
+ struct ceph_cap_flush *cf;
+
+ pr_info("%s: still waiting for cap flushes through %llu\n:\n",
+ __func__, want_tid);
+ spin_lock(&mdsc->cap_dirty_lock);
+ list_for_each_entry(cf, &mdsc->cap_flush_list, g_list) {
+ if (cf->tid > want_tid)
+ break;
+ pr_info("%d: %llu %s %d\n", index++, cf->tid,
+ ceph_cap_string(cf->caps), cf->wake);
+ }
+ spin_unlock(&mdsc->cap_dirty_lock);
+}
+
/*
* flush all dirty inode data to disk.
*
@@ -2072,10 +2089,19 @@ static int check_caps_flush(struct ceph_mds_client *mdsc,
static void wait_caps_flush(struct ceph_mds_client *mdsc,
u64 want_flush_tid)
{
+ long ret;
+
dout("check_caps_flush want %llu\n", want_flush_tid);
- wait_event(mdsc->cap_flushing_wq,
- check_caps_flush(mdsc, want_flush_tid));
+ do {
+ ret = wait_event_timeout(mdsc->cap_flushing_wq,
+ check_caps_flush(mdsc, want_flush_tid), 60 * HZ);
+ if (ret == 0)
+ dump_cap_flushes(mdsc, want_flush_tid);
+ else if (ret == 1)
+ pr_info("%s: condition evaluated to true after timeout!\n",
+ __func__);
+ } while (ret == 0);
dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
}
We've had some cases of hung umounts in teuthology testing. It looks like client is waiting for cap flushes to complete, but they aren't. Change wait_caps_flush to wait 60s, and then dump info about the condition of the list after that point. Reported-by: Patrick Donnelly <pdonnell@redhat.com> URL: https://tracker.ceph.com/issues/51279 Signed-off-by: Jeff Layton <jlayton@kernel.org> --- fs/ceph/mds_client.c | 30 ++++++++++++++++++++++++++++-- 1 file changed, 28 insertions(+), 2 deletions(-) I'm planning to drop this into the testing kernel to help us track down the cause. I'm not sure if we'll want to keep it long term so I'll plan to add a [DO NOT MERGE] tag when I do.