Skip to contentSkip to conversation

nvme: scan sequentially only when list scan unsupported

PATCH In review Uday Shankar 5 participants 7 messages Nov 7, 2022 – Nov 9, 2022 NVMe

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 nvme: scan sequentially only when list scan unsupported

Currently, if nvme_scan_ns_list fails, nvme_scan_work will fall back to a sequential scan. nvme_scan_ns_list can fail for a variety of reasons, e.g. transient transport issue. And the resulting sequential scan can be extremely expensive on controllers reporting an NN value close to the maximum allowed (>4 billion). Avoid sequential scans wherever possible by only falling back to them if nvme_scan_ns_list fails due to controller non-support of Identify NS List. This breaks (noncompliant) devices that claim to support version NVME_VS(1, 1, 0) or later, but don't support Identify NS List. Such devices can be made to work again using the existing NVME_QUIRK_IDENTIFY_CNS.

---
Changes from v1:
- Move limited_cns check from nvme_scan_ns_list to nvme_scan_work
- Move note about devices that may break with this change into commit message

 drivers/nvme/host/core.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/nvme/host/core.c +3 -4
index 0090dc0b3ae6..5abd8d4c6d9b 100644
@@ -4425,9 +4425,6 @@ static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
 	u32 prev = 0;
 	int ret = 0, i;
 
-	if (nvme_ctrl_limited_cns(ctrl))
-		return -EOPNOTSUPP;
-
 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
 	if (!ns_list)
 		return -ENOMEM;
@@ -4535,8 +4532,10 @@ static void nvme_scan_work(struct work_struct *work)
 	}
 
 	mutex_lock(&ctrl->scan_lock);
-	if (nvme_scan_ns_list(ctrl) != 0)
+	if (nvme_ctrl_limited_cns(ctrl))
 		nvme_scan_ns_sequential(ctrl);
+	else
+		nvme_scan_ns_list(ctrl);
 	mutex_unlock(&ctrl->scan_lock);
 }
 

base-commit: d30a909f9bb5283e701a5fdfffac763ef57a3e7c

Signature · 1 line
2.25.1
#1 lore raw

nvme_ctrl_limited_cns check in nvme_scan_ns_list can be removed.

Best regards.

Guixin Liu

Quoted 49 lines · Uday Shankar

Currently, if nvme_scan_ns_list fails, nvme_scan_work will fall back to a sequential scan. nvme_scan_ns_list can fail for a variety of reasons, e.g. transient transport issue. And the resulting sequential scan can be extremely expensive on controllers reporting an NN value close to the maximum allowed (>4 billion). Avoid sequential scans wherever possible by only falling back to them if nvme_scan_ns_list fails due to controller non-support of Identify NS List. This breaks (noncompliant) devices that claim to support version NVME_VS(1, 1, 0) or later, but don't support Identify NS List. Such devices can be made to work again using the existing NVME_QUIRK_IDENTIFY_CNS.

---
Changes from v1:
- Move limited_cns check from nvme_scan_ns_list to nvme_scan_work
- Move note about devices that may break with this change into commit message

  drivers/nvme/host/core.c | 7 +++----
  1 file changed, 3 insertions(+), 4 deletions(-)
drivers/nvme/host/core.c +3 -4
index 0090dc0b3ae6..5abd8d4c6d9b 100644
@@ -4425,9 +4425,6 @@ static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
  	u32 prev = 0;
  	int ret = 0, i;
 
-	if (nvme_ctrl_limited_cns(ctrl))
-		return -EOPNOTSUPP;
-
  	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
  	if (!ns_list)
  		return -ENOMEM;
@@ -4535,8 +4532,10 @@ static void nvme_scan_work(struct work_struct *work)
  	}
 
  	mutex_lock(&ctrl->scan_lock);
-	if (nvme_scan_ns_list(ctrl) != 0)
+	if (nvme_ctrl_limited_cns(ctrl))
  		nvme_scan_ns_sequential(ctrl);
+	else
+		nvme_scan_ns_list(ctrl);
  	mutex_unlock(&ctrl->scan_lock);
  }
 

base-commit: d30a909f9bb5283e701a5fdfffac763ef57a3e7c

#2 lore raw
Quoted 2 lines · Guixin Liu

nvme_ctrl_limited_cns check in nvme_scan_ns_list can be removed.

Sorry, my bad, already removed.

Quoted 54 lines

Best regards.

Guixin Liu

Quoted 48 lines · Uday Shankar

Currently, if nvme_scan_ns_list fails, nvme_scan_work will fall back to a sequential scan. nvme_scan_ns_list can fail for a variety of reasons, e.g. transient transport issue. And the resulting sequential scan can be extremely expensive on controllers reporting an NN value close to the maximum allowed (>4 billion). Avoid sequential scans wherever possible by only falling back to them if nvme_scan_ns_list fails due to controller non-support of Identify NS List. This breaks (noncompliant) devices that claim to support version NVME_VS(1, 1, 0) or later, but don't support Identify NS List. Such devices can be made to work again using the existing NVME_QUIRK_IDENTIFY_CNS.

---
Changes from v1:
- Move limited_cns check from nvme_scan_ns_list to nvme_scan_work
- Move note about devices that may break with this change into commit message

  drivers/nvme/host/core.c | 7 +++----

  1 file changed, 3 insertions(+), 4 deletions(-)
drivers/nvme/host/core.c +0 -2
index 0090dc0b3ae6..5abd8d4c6d9b 100644
@@ -4425,9 +4425,6 @@ static int nvme_scan_ns_list(struct nvme_ctrl 
 *ctrl)
       u32 prev = 0;
       int ret = 0, i;
   -    if (nvme_ctrl_limited_cns(ctrl))
-        return -EOPNOTSUPP;
-
       ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
       if (!ns_list)

          return -ENOMEM;

drivers/nvme/host/core.c +3 -1
@@ -4535,8 +4532,10 @@ static void nvme_scan_work(struct work_struct 
 *work)
       }
         mutex_lock(&ctrl->scan_lock);
-    if (nvme_scan_ns_list(ctrl) != 0)
+    if (nvme_ctrl_limited_cns(ctrl))
           nvme_scan_ns_sequential(ctrl);
+    else
+        nvme_scan_ns_list(ctrl);
       mutex_unlock(&ctrl->scan_lock);
   }
 

base-commit: d30a909f9bb5283e701a5fdfffac763ef57a3e7c

#3 lore raw
In reply to Uday Shankar

Uday,

Quoted 3 lines · Uday Shankar

Currently, if nvme_scan_ns_list fails, nvme_scan_work will fall back to a sequential scan. nvme_scan_ns_list can fail for a variety of reasons, e.g. transient transport issue. And the resulting sequential scan can be

's/issues .And/ issue and/' ?

Quoted 35 lines

extremely expensive on controllers reporting an NN value close to the maximum allowed (>4 billion). Avoid sequential scans wherever possible by only falling back to them if nvme_scan_ns_list fails due to controller non-support of Identify NS List. This breaks (noncompliant) devices that claim to support version NVME_VS(1, 1, 0) or later, but don't support Identify NS List. Such devices can be made to work again using the existing NVME_QUIRK_IDENTIFY_CNS.

---
Changes from v1:
- Move limited_cns check from nvme_scan_ns_list to nvme_scan_work
- Move note about devices that may break with this change into commit message

  drivers/nvme/host/core.c | 7 +++----
  1 file changed, 3 insertions(+), 4 deletions(-)
drivers/nvme/host/core.c +0 -3
index 0090dc0b3ae6..5abd8d4c6d9b 100644
@@ -4425,9 +4425,6 @@ static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
  	u32 prev = 0;
  	int ret = 0, i;
 
-	if (nvme_ctrl_limited_cns(ctrl))
-		return -EOPNOTSUPP;
-
  	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
  	if (!ns_list)
  		return -ENOMEM;
@@ -4535,8 +4532,10 @@ static void nvme_scan_work(struct work_struct *work)
  	}

drivers/nvme/host/core.c static void nvme_scan_work(struct work_struct *work) · line 4541

Quoted 10 lines
drivers/nvme/host/core.c +3 -1
 
  	mutex_lock(&ctrl->scan_lock);
-	if (nvme_scan_ns_list(ctrl) != 0)
+	if (nvme_ctrl_limited_cns(ctrl))
  		nvme_scan_ns_sequential(ctrl);
+	else
+		nvme_scan_ns_list(ctrl);
  	mutex_unlock(&ctrl->scan_lock);
  }
 

Overall it looks good, but I'm curious to know what kind of testing is done on this patch in order to validate the correctness of this patch ? or it this scenario is covered by any blktests/nvme category ?

-ck

maintainer
#4 lore raw
In reply to Uday Shankar
Quoted 10 lines · Uday Shankar

Currently, if nvme_scan_ns_list fails, nvme_scan_work will fall back to a sequential scan. nvme_scan_ns_list can fail for a variety of reasons, e.g. transient transport issue. And the resulting sequential scan can be extremely expensive on controllers reporting an NN value close to the maximum allowed (>4 billion). Avoid sequential scans wherever possible by only falling back to them if nvme_scan_ns_list fails due to controller non-support of Identify NS List. This breaks (noncompliant) devices that claim to support version NVME_VS(1, 1, 0) or later, but don't support Identify NS List. Such devices can be made to work again using the existing NVME_QUIRK_IDENTIFY_CNS.

Looks fine, but I really don't look forward to finding out how many devices, if any, were secretly relying on this fallback.

#5 lore raw
Quoted 2 lines · Keith Busch

Looks fine, but I really don't look forward to finding out how many devices, if any, were secretly relying on this fallback.

I'm a little worried as well. Maybe we can whitelist a few status codes as suggestions that a device might just be a broken. I.e. "NVME_INVALID_FIELD | NVME_DNR" would very much be an indicator of tat.

Also Uday, the return value from nvme_scan_ns_list can and should be dropped now.

#6 lore raw
Quoted 7 lines · Christoph Hellwig
Quoted 2 lines · Keith Busch

Looks fine, but I really don't look forward to finding out how many devices, if any, were secretly relying on this fallback.

I'm a little worried as well. Maybe we can whitelist a few status codes as suggestions that a device might just be a broken. I.e. "NVME_INVALID_FIELD | NVME_DNR" would very much be an indicator of tat.

Can we get away with just testing for the DNR bit? The practical goal of this patch is to prevent fallback to sequential scan when the Identify NS List command fails for transient reasons, and I think the error code returned in such cases should not have DNR set.

Quoted 2 lines

Also Uday, the return value from nvme_scan_ns_list can and should be dropped now.

We'll want to keep the return value if we go with the above suggestion.