nvme-tcp: fix potential unbalanced freeze & unfreeze
authorMing Lei <ming.lei@redhat.com>
Tue, 11 Jul 2023 09:40:40 +0000 (17:40 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 16 Aug 2023 16:27:30 +0000 (18:27 +0200)
commit 99dc264014d5aed66ee37ddf136a38b5a2b1b529 upstream.

Move start_freeze into nvme_tcp_configure_io_queues(), and there is
at least two benefits:

1) fix unbalanced freeze and unfreeze, since re-connection work may
fail or be broken by removal

2) IO during error recovery can be failfast quickly because nvme fabrics
unquiesces queues after teardown.

One side-effect is that !mpath request may timeout during connecting
because of queue topo change, but that looks not one big deal:

1) same problem exists with current code base

2) compared with !mpath, mpath use case is dominant

Fixes: 2875b0aecabe ("nvme-tcp: fix controller reset hang during traffic")
Cc: stable@vger.kernel.org
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Tested-by: Yi Zhang <yi.zhang@redhat.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/nvme/host/tcp.c

index 8f17cbe..f2fedd2 100644 (file)
@@ -1890,6 +1890,7 @@ static int nvme_tcp_configure_io_queues(struct nvme_ctrl *ctrl, bool new)
                goto out_cleanup_connect_q;
 
        if (!new) {
+               nvme_start_freeze(ctrl);
                nvme_start_queues(ctrl);
                if (!nvme_wait_freeze_timeout(ctrl, NVME_IO_TIMEOUT)) {
                        /*
@@ -1898,6 +1899,7 @@ static int nvme_tcp_configure_io_queues(struct nvme_ctrl *ctrl, bool new)
                         * to be safe.
                         */
                        ret = -ENODEV;
+                       nvme_unfreeze(ctrl);
                        goto out_wait_freeze_timed_out;
                }
                blk_mq_update_nr_hw_queues(ctrl->tagset,
@@ -2002,7 +2004,6 @@ static void nvme_tcp_teardown_io_queues(struct nvme_ctrl *ctrl,
        if (ctrl->queue_count <= 1)
                return;
        nvme_stop_admin_queue(ctrl);
-       nvme_start_freeze(ctrl);
        nvme_stop_queues(ctrl);
        nvme_sync_io_queues(ctrl);
        nvme_tcp_stop_io_queues(ctrl);