There is a race between iterating over requests in
nbd_clear_que() and completing requests in recv_work(),
which can lead to double completion of a request.
To fix it, flush the recv worker before iterating over
the requests and don't abort the completed request
while iterating.
Fixes: 96d97e17828f ("nbd: clear_sock on netlink disconnect")
Reported-by: Jiang Yadong <jiangyadong@xxxxxxxxxxxxx>
Signed-off-by: Xie Yongji <xieyongji@xxxxxxxxxxxxx>
---
drivers/block/nbd.c | 14 +++++++++++---
1 file changed, 11 insertions(+), 3 deletions(-)
diff --git a/drivers/block/nbd.c b/drivers/block/nbd.c
index c38317979f74..19f5d5a8b16a 100644
--- a/drivers/block/nbd.c
+++ b/drivers/block/nbd.c
@@ -818,6 +818,10 @@ static bool nbd_clear_req(struct request *req, void *data, bool reserved)
{
struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
+ /* don't abort one completed request */
+ if (blk_mq_request_completed(req))
+ return true;
+
mutex_lock(&cmd->lock);
cmd->status = BLK_STS_IOERR;
mutex_unlock(&cmd->lock);
@@ -2004,15 +2008,19 @@ static void nbd_disconnect_and_put(struct nbd_device *nbd)
{
mutex_lock(&nbd->config_lock);
nbd_disconnect(nbd);
- nbd_clear_sock(nbd);
- mutex_unlock(&nbd->config_lock);
+ sock_shutdown(nbd);
/*
* Make sure recv thread has finished, so it does not drop the last
* config ref and try to destroy the workqueue from inside the work
- * queue.
+ * queue. And this also ensure that we can safely call nbd_clear_que()
+ * to cancel the inflight I/Os.
*/
if (nbd->recv_workq)
flush_workqueue(nbd->recv_workq);
+ nbd_clear_que(nbd);
+ nbd->task_setup = NULL;
+ mutex_unlock(&nbd->config_lock);
+
if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
&nbd->config->runtime_flags))
nbd_config_put(nbd);
--
2.11.0
