[PATCHv2 0/2] nvme: short-circuit connection retries

[PATCHv2 0/2] nvme: short-circuit connection retries

[Hannes Reinecke]

Hi all,

here are two patches updating nvme-tcp and nvme-rdma transport
to modifications we've already have in nvme-fc with commit
f25f8ef70ce2 ("nvme-fc: short-circuit reconnect retries").
Gist is to short-circuit reconnection retries if we get a
status back with the DNR bit set, in which case we shouldn't
retry the connection attempt.

As usual, comments and reviews are welcome.

Changes to the original version:
- Rework with suggestions from Sagi

Hannes Reinecke (2):
  nvme-tcp: short-circuit connect retries
  nvme-rdma: short-circuit connect retries

 drivers/nvme/host/rdma.c | 20 +++++++++++++++++---
 drivers/nvme/host/tcp.c  | 21 ++++++++++++++++++---
 2 files changed, 35 insertions(+), 6 deletions(-)

-- 
2.29.2

[PATCH 1/2] nvme-tcp: short-circuit connect retries

[Hannes Reinecke]

When a reconnect attempt fails with a non-retryable status
(eg when the subsystem has been unprovisioned) there hardly
is any reason to retry the reconnect attempt.
So check the actual error status from nvme_tcp_setup_ctrl(), and
call nvme_delete_ctrl() instead of calling nvme_tcp_reconnect_or_remove()
if the DNR bit is set.

Signed-off-by: Hannes Reinecke <hare@suse.de>
---
 drivers/nvme/host/tcp.c | 21 ++++++++++++++++++---
 1 file changed, 18 insertions(+), 3 deletions(-)

diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
index c854e69defb0..a15e3ed02b64 100644
--- a/drivers/nvme/host/tcp.c
+++ b/drivers/nvme/host/tcp.c
@@ -2157,10 +2157,12 @@ static void nvme_tcp_reconnect_ctrl_work(struct work_struct *work)
 	struct nvme_tcp_ctrl *tcp_ctrl = container_of(to_delayed_work(work),
 			struct nvme_tcp_ctrl, connect_work);
 	struct nvme_ctrl *ctrl = &tcp_ctrl->ctrl;
+	int ret;
 
 	++ctrl->nr_reconnects;
 
-	if (nvme_tcp_setup_ctrl(ctrl, false))
+	ret = nvme_tcp_setup_ctrl(ctrl, false);
+	if (ret)
 		goto requeue;
 
 	dev_info(ctrl->device, "Successfully reconnected (%d attempt)\n",
@@ -2173,6 +2175,12 @@ static void nvme_tcp_reconnect_ctrl_work(struct work_struct *work)
 requeue:
 	dev_info(ctrl->device, "Failed reconnect attempt %d\n",
 			ctrl->nr_reconnects);
+	if (ret > 0 && (ret & NVME_SC_DNR)) {
+		dev_info(ctrl->device,
+			 "Removing ctrl...\n");
+		nvme_delete_ctrl(ctrl);
+		return;
+	}
 	nvme_tcp_reconnect_or_remove(ctrl);
 }
 
@@ -2224,6 +2232,7 @@ static void nvme_reset_ctrl_work(struct work_struct *work)
 {
 	struct nvme_ctrl *ctrl =
 		container_of(work, struct nvme_ctrl, reset_work);
+	int ret;
 
 	nvme_stop_ctrl(ctrl);
 	nvme_tcp_teardown_ctrl(ctrl, false);
@@ -2235,9 +2244,15 @@ static void nvme_reset_ctrl_work(struct work_struct *work)
 		return;
 	}
 
-	if (nvme_tcp_setup_ctrl(ctrl, false))
+	ret = nvme_tcp_setup_ctrl(ctrl, false);
+	if (ret) {
+		if (ret > 0 && (ret & NVME_SC_DNR)) {
+			dev_info(ctrl->device, "Removing ctrl...\n");
+			nvme_delete_ctrl(ctrl);
+			return;
+		}
 		goto out_fail;
-
+	}
 	return;
 
 out_fail:
-- 
2.29.2

[PATCH 2/2] nvme-rdma: short-circuit connect retries

[Hannes Reinecke]

When a reconnect attempt fails with a non-retryable status
(eg when the subsystem has been unprovisioned) there hardly
is any reason to retry the reconnect attempt.
So check the actual error status from nvme_rdma_setup_ctrl(), and
call nvme_delete_ctrl() instead of calling nvme_rdma_reconnect_or_remove()
if the DNR bit is set.

Signed-off-by: Hannes Reinecke <hare@suse.de>
---
 drivers/nvme/host/rdma.c | 20 +++++++++++++++++---
 1 file changed, 17 insertions(+), 3 deletions(-)

diff --git a/drivers/nvme/host/rdma.c b/drivers/nvme/host/rdma.c
index 7d01fb770284..a43e47c93140 100644
--- a/drivers/nvme/host/rdma.c
+++ b/drivers/nvme/host/rdma.c
@@ -1170,10 +1170,12 @@ static void nvme_rdma_reconnect_ctrl_work(struct work_struct *work)
 {
 	struct nvme_rdma_ctrl *ctrl = container_of(to_delayed_work(work),
 			struct nvme_rdma_ctrl, reconnect_work);
+	int ret;
 
 	++ctrl->ctrl.nr_reconnects;
 
-	if (nvme_rdma_setup_ctrl(ctrl, false))
+	ret = nvme_rdma_setup_ctrl(ctrl, false);
+	if (ret)
 		goto requeue;
 
 	dev_info(ctrl->ctrl.device, "Successfully reconnected (%d attempts)\n",
@@ -1186,6 +1188,11 @@ static void nvme_rdma_reconnect_ctrl_work(struct work_struct *work)
 requeue:
 	dev_info(ctrl->ctrl.device, "Failed reconnect attempt %d\n",
 			ctrl->ctrl.nr_reconnects);
+	if (ret > 0 && (ret & NVME_SC_DNR)) {
+		dev_info(ctrl->ctrl.device, "Removing ctrl...\n");
+		nvme_delete_ctrl(&ctrl->ctrl);
+		return;
+	}
 	nvme_rdma_reconnect_or_remove(ctrl);
 }
 
@@ -2271,6 +2278,7 @@ static void nvme_rdma_reset_ctrl_work(struct work_struct *work)
 {
 	struct nvme_rdma_ctrl *ctrl =
 		container_of(work, struct nvme_rdma_ctrl, ctrl.reset_work);
+	int ret;
 
 	nvme_stop_ctrl(&ctrl->ctrl);
 	nvme_rdma_shutdown_ctrl(ctrl, false);
@@ -2281,9 +2289,15 @@ static void nvme_rdma_reset_ctrl_work(struct work_struct *work)
 		return;
 	}
 
-	if (nvme_rdma_setup_ctrl(ctrl, false))
+	ret = nvme_rdma_setup_ctrl(ctrl, false);
+	if (ret) {
+		if (ret > 0 && (ret & NVME_SC_DNR)) {
+			dev_info(ctrl->ctrl.device, "Removing ctrl...\n");
+			nvme_delete_ctrl(&ctrl->ctrl);
+			return;
+		}
 		goto out_fail;
-
+	}
 	return;
 
 out_fail:
-- 
2.29.2

[PATCH 0/2] nvme: short-circuit connection retries

[Hannes Reinecke]

Hi all,

here are two patches updating nvme-tcp and nvme-rdma transport
to modifications we've already have in nvme-fc with commit
f25f8ef70ce2 ("nvme-fc: short-circuit reconnect retries").
Gist is to short-circuit reconnection retries if we get a
status back with the DNR bit set, in which case we shouldn't
retry the connection attempt.

As usual, comments and reviews are welcome.

Hannes Reinecke (2):
  nvme-tcp: short-circuit connect retries
  nvme-rdma: short-circuit connect retries

 drivers/nvme/host/rdma.c | 25 ++++++++++++++++++-------
 drivers/nvme/host/tcp.c  | 25 ++++++++++++++++++-------
 2 files changed, 36 insertions(+), 14 deletions(-)

-- 
2.29.2

[PATCH 2/2] nvme-rdma: short-circuit connect retries

[Hannes Reinecke]

When a reconnect attempt fails with a non-retryable status
(eg when the subsystem has been unprovisioned) there hardly
is any reason to retry the reconnect attempt.
So pass the actual error status to nvme_tcp_reconnect_or_remove()
and short-circuit retries if the DNR bit is set.

Signed-off-by: Hannes Reinecke <hare@suse.de>
---
 drivers/nvme/host/rdma.c | 25 ++++++++++++++++++-------
 1 file changed, 18 insertions(+), 7 deletions(-)

diff --git a/drivers/nvme/host/rdma.c b/drivers/nvme/host/rdma.c
index 84ce3347d158..bcc84f181dcd 100644
--- a/drivers/nvme/host/rdma.c
+++ b/drivers/nvme/host/rdma.c
@@ -1065,8 +1065,10 @@ static void nvme_rdma_free_ctrl(struct nvme_ctrl *nctrl)
 	kfree(ctrl);
 }
 
-static void nvme_rdma_reconnect_or_remove(struct nvme_rdma_ctrl *ctrl)
+static void nvme_rdma_reconnect_or_remove(struct nvme_rdma_ctrl *ctrl, int status)
 {
+	bool recon = true;
+
 	/* If we are resetting/deleting then do nothing */
 	if (ctrl->ctrl.state != NVME_CTRL_CONNECTING) {
 		WARN_ON_ONCE(ctrl->ctrl.state == NVME_CTRL_NEW ||
@@ -1074,7 +1076,12 @@ static void nvme_rdma_reconnect_or_remove(struct nvme_rdma_ctrl *ctrl)
 		return;
 	}
 
-	if (nvmf_should_reconnect(&ctrl->ctrl)) {
+	if (status > 0 && (status & NVME_SC_DNR)) {
+		dev_info(ctrl->ctrl.device, "reconnect failure %d\n", status);
+		recon = false;
+	}
+
+	if (recon && nvmf_should_reconnect(&ctrl->ctrl)) {
 		dev_info(ctrl->ctrl.device, "Reconnecting in %d seconds...\n",
 			ctrl->ctrl.opts->reconnect_delay);
 		queue_delayed_work(nvme_wq, &ctrl->reconnect_work,
@@ -1173,10 +1180,12 @@ static void nvme_rdma_reconnect_ctrl_work(struct work_struct *work)
 {
 	struct nvme_rdma_ctrl *ctrl = container_of(to_delayed_work(work),
 			struct nvme_rdma_ctrl, reconnect_work);
+	int ret;
 
 	++ctrl->ctrl.nr_reconnects;
 
-	if (nvme_rdma_setup_ctrl(ctrl, false))
+	ret = nvme_rdma_setup_ctrl(ctrl, false);
+	if (ret)
 		goto requeue;
 
 	dev_info(ctrl->ctrl.device, "Successfully reconnected (%d attempts)\n",
@@ -1189,7 +1198,7 @@ static void nvme_rdma_reconnect_ctrl_work(struct work_struct *work)
 requeue:
 	dev_info(ctrl->ctrl.device, "Failed reconnect attempt %d\n",
 			ctrl->ctrl.nr_reconnects);
-	nvme_rdma_reconnect_or_remove(ctrl);
+	nvme_rdma_reconnect_or_remove(ctrl, ret);
 }
 
 static void nvme_rdma_error_recovery_work(struct work_struct *work)
@@ -1212,7 +1221,7 @@ static void nvme_rdma_error_recovery_work(struct work_struct *work)
 		return;
 	}
 
-	nvme_rdma_reconnect_or_remove(ctrl);
+	nvme_rdma_reconnect_or_remove(ctrl, -ENOTCONN);
 }
 
 static void nvme_rdma_error_recovery(struct nvme_rdma_ctrl *ctrl)
@@ -2274,6 +2283,7 @@ static void nvme_rdma_reset_ctrl_work(struct work_struct *work)
 {
 	struct nvme_rdma_ctrl *ctrl =
 		container_of(work, struct nvme_rdma_ctrl, ctrl.reset_work);
+	int ret;
 
 	nvme_stop_ctrl(&ctrl->ctrl);
 	nvme_rdma_shutdown_ctrl(ctrl, false);
@@ -2284,14 +2294,15 @@ static void nvme_rdma_reset_ctrl_work(struct work_struct *work)
 		return;
 	}
 
-	if (nvme_rdma_setup_ctrl(ctrl, false))
+	ret = nvme_rdma_setup_ctrl(ctrl, false);
+	if (ret)
 		goto out_fail;
 
 	return;
 
 out_fail:
 	++ctrl->ctrl.nr_reconnects;
-	nvme_rdma_reconnect_or_remove(ctrl);
+	nvme_rdma_reconnect_or_remove(ctrl, ret);
 }
 
 static const struct nvme_ctrl_ops nvme_rdma_ctrl_ops = {
-- 
2.29.2