rbd: don't wait for the lock forever if blacklisted
-EBLACKLISTED from __rbd_register_watch() means that our ceph_client
got blacklisted - we won't be able to restore the watch and reacquire
the lock. Wake up and fail all outstanding requests waiting for the
lock and arrange for all new requests that require the lock to fail
immediately.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Tested-by: Mike Christie <mchristi@redhat.com>
diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c
index abb7162..633e8c2 100644
--- a/drivers/block/rbd.c
+++ b/drivers/block/rbd.c
@@ -415,15 +415,15 @@
};
/*
- * Flag bits for rbd_dev->flags. If atomicity is required,
- * rbd_dev->lock is used to protect access.
- *
- * Currently, only the "removing" flag (which is coupled with the
- * "open_count" field) requires atomic access.
+ * Flag bits for rbd_dev->flags:
+ * - REMOVING (which is coupled with rbd_dev->open_count) is protected
+ * by rbd_dev->lock
+ * - BLACKLISTED is protected by rbd_dev->lock_rwsem
*/
enum rbd_dev_flags {
RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */
RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */
+ RBD_DEV_FLAG_BLACKLISTED, /* our ceph_client is blacklisted */
};
static DEFINE_MUTEX(client_mutex); /* Serialize client creation */
@@ -3926,6 +3926,7 @@
struct rbd_device *rbd_dev = container_of(to_delayed_work(work),
struct rbd_device, watch_dwork);
bool was_lock_owner = false;
+ bool need_to_wake = false;
int ret;
dout("%s rbd_dev %p\n", __func__, rbd_dev);
@@ -3935,19 +3936,27 @@
was_lock_owner = rbd_release_lock(rbd_dev);
mutex_lock(&rbd_dev->watch_mutex);
- if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR)
- goto fail_unlock;
+ if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) {
+ mutex_unlock(&rbd_dev->watch_mutex);
+ goto out;
+ }
ret = __rbd_register_watch(rbd_dev);
if (ret) {
rbd_warn(rbd_dev, "failed to reregister watch: %d", ret);
- if (ret != -EBLACKLISTED)
+ if (ret == -EBLACKLISTED) {
+ set_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags);
+ need_to_wake = true;
+ } else {
queue_delayed_work(rbd_dev->task_wq,
&rbd_dev->watch_dwork,
RBD_RETRY_DELAY);
- goto fail_unlock;
+ }
+ mutex_unlock(&rbd_dev->watch_mutex);
+ goto out;
}
+ need_to_wake = true;
rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED;
rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id;
mutex_unlock(&rbd_dev->watch_mutex);
@@ -3963,13 +3972,10 @@
ret);
}
+out:
up_write(&rbd_dev->lock_rwsem);
- wake_requests(rbd_dev, true);
- return;
-
-fail_unlock:
- mutex_unlock(&rbd_dev->watch_mutex);
- up_write(&rbd_dev->lock_rwsem);
+ if (need_to_wake)
+ wake_requests(rbd_dev, true);
}
/*
@@ -4074,7 +4080,9 @@
up_read(&rbd_dev->lock_rwsem);
schedule();
down_read(&rbd_dev->lock_rwsem);
- } while (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED);
+ } while (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED &&
+ !test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags));
+
finish_wait(&rbd_dev->lock_waitq, &wait);
}
@@ -4166,8 +4174,16 @@
if (must_be_locked) {
down_read(&rbd_dev->lock_rwsem);
- if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED)
+ if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED &&
+ !test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags))
rbd_wait_state_locked(rbd_dev);
+
+ WARN_ON((rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) ^
+ !test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags));
+ if (test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags)) {
+ result = -EBLACKLISTED;
+ goto err_unlock;
+ }
}
img_request = rbd_img_request_create(rbd_dev, offset, length, op_type,