[PATCH v3 0/4] nvme: improve error handling and ana_state to work well with dm-multipath

[PATCH v3 0/4] nvme: improve error handling and ana_state to work well with dm-multipath

[Mike Snitzer]

Hi,

This patchset reflects changes needed to make NVMe error handling and
ANA state updates work well with dm-multipath (which always sets
REQ_FAILFAST_TRANSPORT).

RHEL8 has been carrying an older ~5.9 based version of this patchset
(since RHEL8.3, August 2020).

RHEL9 is coming, would really prefer that these changes land upstream
rather than carry them within RHEL.

All review/feedback welcome.

Thanks,
Mike

v2 -> v3:
- Added Reviewed-by tags to BLK_STS_DO_NOT_RETRY patch.
- Eliminated __nvme_end_req() and added code comment to
  nvme_failup_req() in FAILUP handling patch.

Chao Leng (1):
  nvme: allow local retry for requests with REQ_FAILFAST_TRANSPORT set

Mike Snitzer (3):
  nvme: return BLK_STS_DO_NOT_RETRY if the DNR bit is set
  nvme: introduce FAILUP handling for REQ_FAILFAST_TRANSPORT
  nvme: decouple basic ANA log page re-read support from native
    multipathing

 drivers/nvme/host/core.c      | 42 +++++++++++++++++++++++++++++++++++++++---
 drivers/nvme/host/multipath.c | 16 +++++++++++-----
 drivers/nvme/host/nvme.h      |  4 ++++
 include/linux/blk_types.h     |  8 ++++++++
 4 files changed, 62 insertions(+), 8 deletions(-)

-- 
2.15.0

[PATCH v3 1/4] nvme: return BLK_STS_DO_NOT_RETRY if the DNR bit is set

[Mike Snitzer]

If the DNR bit is set we should not retry the command.

We care about the retryable vs not retryable distinction at the block
layer so propagate the equivalent of the DNR bit by introducing
BLK_STS_DO_NOT_RETRY. Update blk_path_error() to _not_ retry if it
is set.

This change runs with the suggestion made here:
https://lore.kernel.org/linux-nvme/20190813170144.GA10269@lst.de/

Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
---
 drivers/nvme/host/core.c  | 3 +++
 include/linux/blk_types.h | 8 ++++++++
 2 files changed, 11 insertions(+)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 0896e21642be..540d6fd8ffef 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -237,6 +237,9 @@ static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 
 static blk_status_t nvme_error_status(u16 status)
 {
+	if (unlikely(status & NVME_SC_DNR))
+		return BLK_STS_DO_NOT_RETRY;
+
 	switch (status & 0x7ff) {
 	case NVME_SC_SUCCESS:
 		return BLK_STS_OK;
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index db026b6ec15a..1ca724948c56 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -142,6 +142,13 @@ typedef u8 __bitwise blk_status_t;
  */
 #define BLK_STS_ZONE_ACTIVE_RESOURCE	((__force blk_status_t)16)
 
+/*
+ * BLK_STS_DO_NOT_RETRY is returned from the driver in the completion path
+ * if the device returns a status indicating that if the same command is
+ * re-submitted it is expected to fail.
+ */
+#define BLK_STS_DO_NOT_RETRY	((__force blk_status_t)17)
+
 /**
  * blk_path_error - returns true if error may be path related
  * @error: status the request was completed with
@@ -157,6 +164,7 @@ typedef u8 __bitwise blk_status_t;
 static inline bool blk_path_error(blk_status_t error)
 {
 	switch (error) {
+	case BLK_STS_DO_NOT_RETRY:
 	case BLK_STS_NOTSUPP:
 	case BLK_STS_NOSPC:
 	case BLK_STS_TARGET:
-- 
2.15.0

[PATCH v3 2/4] nvme: allow local retry for requests with REQ_FAILFAST_TRANSPORT set

[Mike Snitzer]

From: Chao Leng <lengchao@huawei.com>

REQ_FAILFAST_TRANSPORT was designed for SCSI, because the SCSI protocol
does not define the local retry mechanism. SCSI implements a fuzzy
local retry mechanism, so REQ_FAILFAST_TRANSPORT is needed to allow
higher-level multipathing software to perform failover/retry.

NVMe is different with SCSI about this. It defines a local retry
mechanism and path error codes, so NVMe should retry local for non
path error. If path related error, whether to retry and how to retry
is still determined by higher-level multipathing's failover.

Unlike SCSI, NVMe shouldn't prevent retry if REQ_FAILFAST_TRANSPORT
because NVMe's local retry is needed -- as is NVMe specific logic to
categorize whether an error is path related.

In this way, the mechanism of NVMe multipath or other multipath are
now equivalent. The mechanism is: non path related error will be
retried locally, path related error is handled by multipath.

Signed-off-by: Chao Leng <lengchao@huawei.com>
[snitzer: edited header for grammar and clarity, also added code comment]
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
---
 drivers/nvme/host/core.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 540d6fd8ffef..4134cf3c7e48 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -306,7 +306,14 @@ static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 	if (likely(nvme_req(req)->status == 0))
 		return COMPLETE;
 
-	if (blk_noretry_request(req) ||
+	/*
+	 * REQ_FAILFAST_TRANSPORT is set by upper layer software that
+	 * handles multipathing. Unlike SCSI, NVMe's error handling was
+	 * specifically designed to handle local retry for non-path errors.
+	 * As such, allow NVMe's local retry mechanism to be used for
+	 * requests marked with REQ_FAILFAST_TRANSPORT.
+	 */
+	if ((req->cmd_flags & (REQ_FAILFAST_DEV | REQ_FAILFAST_DRIVER)) ||
 	    (nvme_req(req)->status & NVME_SC_DNR) ||
 	    nvme_req(req)->retries >= nvme_max_retries)
 		return COMPLETE;
-- 
2.15.0

[PATCH v3 3/4] nvme: introduce FAILUP handling for REQ_FAILFAST_TRANSPORT

[Mike Snitzer]

If REQ_FAILFAST_TRANSPORT is set it means the driver should not retry
IO that completed with transport errors. REQ_FAILFAST_TRANSPORT is
set by multipathing software (e.g. dm-multipath) before it issues IO.

Update NVMe to allow failover of requests marked with either
REQ_NVME_MPATH or REQ_FAILFAST_TRANSPORT. This allows such requests
to be given a disposition of either FAILOVER or FAILUP respectively.
FAILUP handling ensures a retryable error is returned up from NVMe.

Introduce nvme_failup_req() for use in nvme_complete_rq() if
nvme_decide_disposition() returns FAILUP. nvme_failup_req() ensures
the request is completed with a retryable IO error when appropriate.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
---
 drivers/nvme/host/core.c | 28 ++++++++++++++++++++++++++--
 1 file changed, 26 insertions(+), 2 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 4134cf3c7e48..605ffba6835f 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -299,6 +299,7 @@ enum nvme_disposition {
 	COMPLETE,
 	RETRY,
 	FAILOVER,
+	FAILUP,
 };
 
 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
@@ -318,10 +319,11 @@ static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 	    nvme_req(req)->retries >= nvme_max_retries)
 		return COMPLETE;
 
-	if (req->cmd_flags & REQ_NVME_MPATH) {
+	if (req->cmd_flags & (REQ_NVME_MPATH | REQ_FAILFAST_TRANSPORT)) {
 		if (nvme_is_path_error(nvme_req(req)->status) ||
 		    blk_queue_dying(req->q))
-			return FAILOVER;
+			return (req->cmd_flags & REQ_NVME_MPATH) ?
+				FAILOVER : FAILUP;
 	} else {
 		if (blk_queue_dying(req->q))
 			return COMPLETE;
@@ -343,6 +345,25 @@ static inline void nvme_end_req(struct request *req)
 	blk_mq_end_request(req, status);
 }
 
+static void nvme_failup_req(struct request *req)
+{
+	blk_status_t status = nvme_error_status(nvme_req(req)->status);
+
+	/* Ensure a retryable path error is returned */
+	if (WARN_ON_ONCE(!blk_path_error(status))) {
+		/*
+		 * If here, nvme_is_path_error() returned true.
+		 * So nvme_error_status() translation needs updating
+		 * relative to blk_path_error(), or vice versa.
+		 */
+		pr_debug("Request meant for failover but blk_status_t (errno=%d) was not retryable.\n",
+			 blk_status_to_errno(status));
+		nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
+	}
+
+	nvme_end_req(req);
+}
+
 void nvme_complete_rq(struct request *req)
 {
 	trace_nvme_complete_rq(req);
@@ -361,6 +382,9 @@ void nvme_complete_rq(struct request *req)
 	case FAILOVER:
 		nvme_failover_req(req);
 		return;
+	case FAILUP:
+		nvme_failup_req(req);
+		return;
 	}
 }
 EXPORT_SYMBOL_GPL(nvme_complete_rq);
-- 
2.15.0

[PATCH v3 4/4] nvme: decouple basic ANA log page re-read support from native multipathing

[Mike Snitzer]

Whether or not ANA is present is a choice of the target implementation;
the host (and whether it supports multipathing) has _zero_ influence on
this. If the target declares a path as 'inaccessible' the path _is_
inaccessible to the host. As such, ANA support should be functional
even if native multipathing is not.

Introduce ability to always re-read ANA log page as required due to ANA
error and make current ANA state available via sysfs -- even if native
multipathing is disabled on the host (e.g. nvme_core.multipath=N).
This is achieved by factoring out nvme_update_ana() and calling it in
nvme_complete_rq() for all FAILOVER requests.

This affords userspace access to the current ANA state independent of
which layer might be doing multipathing. This makes 'nvme list-subsys'
show ANA state for all NVMe subsystems with multiple controllers. It
also allows userspace multipath-tools to rely on the NVMe driver for
ANA support while dm-multipath takes care of multipathing.

And as always, if embedded NVMe users do not want any performance
overhead associated with ANA or native NVMe multipathing they can
disable CONFIG_NVME_MULTIPATH.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
---
 drivers/nvme/host/core.c      |  2 ++
 drivers/nvme/host/multipath.c | 16 +++++++++++-----
 drivers/nvme/host/nvme.h      |  4 ++++
 3 files changed, 17 insertions(+), 5 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 605ffba6835f..9a878a599897 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -349,6 +349,8 @@ static void nvme_failup_req(struct request *req)
 {
 	blk_status_t status = nvme_error_status(nvme_req(req)->status);
 
+	nvme_update_ana(req);
+
 	/* Ensure a retryable path error is returned */
 	if (WARN_ON_ONCE(!blk_path_error(status))) {
 		/*
diff --git a/drivers/nvme/host/multipath.c b/drivers/nvme/host/multipath.c
index a1d476e1ac02..7d94250264aa 100644
--- a/drivers/nvme/host/multipath.c
+++ b/drivers/nvme/host/multipath.c
@@ -65,23 +65,29 @@ void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
 	}
 }
 
-void nvme_failover_req(struct request *req)
+void nvme_update_ana(struct request *req)
 {
 	struct nvme_ns *ns = req->q->queuedata;
 	u16 status = nvme_req(req)->status & 0x7ff;
-	unsigned long flags;
-
-	nvme_mpath_clear_current_path(ns);
 
 	/*
 	 * If we got back an ANA error, we know the controller is alive but not
-	 * ready to serve this namespace.  Kick of a re-read of the ANA
+	 * ready to serve this namespace.  Kick off a re-read of the ANA
 	 * information page, and just try any other available path for now.
 	 */
 	if (nvme_is_ana_error(status) && ns->ctrl->ana_log_buf) {
 		set_bit(NVME_NS_ANA_PENDING, &ns->flags);
 		queue_work(nvme_wq, &ns->ctrl->ana_work);
 	}
+}
+
+void nvme_failover_req(struct request *req)
+{
+	struct nvme_ns *ns = req->q->queuedata;
+	unsigned long flags;
+
+	nvme_mpath_clear_current_path(ns);
+	nvme_update_ana(req);
 
 	spin_lock_irqsave(&ns->head->requeue_lock, flags);
 	blk_steal_bios(&ns->head->requeue_list, req);
diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
index 07b34175c6ce..4eed8536625c 100644
--- a/drivers/nvme/host/nvme.h
+++ b/drivers/nvme/host/nvme.h
@@ -664,6 +664,7 @@ void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
 			struct nvme_ctrl *ctrl, int *flags);
 void nvme_failover_req(struct request *req);
+void nvme_update_ana(struct request *req);
 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
@@ -714,6 +715,9 @@ static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
 static inline void nvme_failover_req(struct request *req)
 {
 }
+static inline void nvme_update_ana(struct request *req)
+{
+}
 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
 {
 }
-- 
2.15.0