Skip to contentSkip to conversation

nvme: correctly account for namespace head reference counter

PATCH Superseded Nilay Shroff 3 participants 8 messages Jun 20, 2025 – Jun 25, 2025 NVMe

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 nvme: correctly account for namespace head reference counter

The blktests nvme/058 menifests an issue where the NVMe subsystem kobject entry remains stale in sysfs, causing a failure during subsequent NVMe module reloads[1]. Specifically, when attempting to register a new NVMe subsystem, the driver encounters a kobejct name collision because a stale kobject still exists. Though, please note that nvme/058 doesn't report any failure and test case passes and it's only during subsequent NVMe module reloads, the stale nvme sub- system kobject entry in sysfs causes the observed symptom[1].

This issue stems from an imbalance in the get/put usage of the namespace head (nshead) reference counter. The nshead holds a reference to the associated NVMe subsystem. If the nshead reference is not properly released, it prevents the cleanup of the subsystem's kobject, leaving nvme subsystem stale entry behind in sysfs.

During the failre case, the last namespace path referencing a nshead is removed, but the nshead reference was not released. This occurs because the release logic currently only puts the nshead reference when its state is LIVE. However, in configurations where ANA (Asymmetric Namespace Access) is enabled, a namespace may be associated with an ANA state that is neither optimized nor non-optimized. In this case, the nshead may never transition to LIVE, and the corresponding nshead reference is then never dropped. In fact nvme/058 associates some of nvme namespaces to an inaccessible ANA state and with that nshead is created but it's state is not transitioned to LIVE. So the current logic would then causes nshead reference to be leaked for non-LIVE states.

Another scenario, during namespace allocation, the driver first allocates a nshead and then issues an Identify Namespace command. If this command fails — which can happen in tests like nvme/058 that rapidly enables and disables namespaces — we must release the reference to the newly allocated nshead. However this reference release is currently missing in the failure, causing a nshead reference leak.

To fix this, we now unconditionally release the nshead reference when the last nvme path referencing to the nshead is removed, regardless of the head’s state. Also during idnetify namespace failure case we now properly release the nshead refernce. So this ensures proper cleanup of the nshead, and consequently, the NVMe subsystem and its associated kobject.

This change prevents stale kobject entries from lingering in sysfs and eliminates the module reload failures observed just after running nvme/058.

[1] https://lore.kernel.org/all/CAHj4cs8fOBS-eSjsd5LUBzy7faKXJtgLkCN+mDy_-ezCLLLq+Q@mail.gmail.com/

Reported-by [email protected]
Fixes 62188639ec16 ("nvme-multipath: introduce delayed removal of the multipath head node")
---
 drivers/nvme/host/core.c      | 7 ++++++-
 drivers/nvme/host/multipath.c | 2 +-
 2 files changed, 7 insertions(+), 2 deletions(-)
drivers/nvme/host/core.c +6 -1
index 92697f98c601..b3e280149a25 100644
@@ -4089,6 +4089,7 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
 	struct nvme_ns *ns;
 	struct gendisk *disk;
 	int node = ctrl->numa_node;
+	bool last_path = false;
 
 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
 	if (!ns)
@@ -4181,9 +4182,13 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
  out_unlink_ns:
 	mutex_lock(&ctrl->subsys->lock);
 	list_del_rcu(&ns->siblings);
-	if (list_empty(&ns->head->list))
+	if (list_empty(&ns->head->list)) {
 		list_del_init(&ns->head->entry);
+		last_path = true;
+	}
 	mutex_unlock(&ctrl->subsys->lock);
+	if (last_path)
+		nvme_put_ns_head(ns->head);
 	nvme_put_ns_head(ns->head);
  out_cleanup_disk:
 	put_disk(disk);
drivers/nvme/host/multipath.c +1 -1
index e040e467f9fa..5f52b50907a0 100644
@@ -690,8 +690,8 @@ static void nvme_remove_head(struct nvme_ns_head *head)
 		nvme_cdev_del(&head->cdev, &head->cdev_device);
 		synchronize_srcu(&head->srcu);
 		del_gendisk(head->disk);
-		nvme_put_ns_head(head);
 	}
+	nvme_put_ns_head(head);
 }
 
 static void nvme_remove_head_work(struct work_struct *work)
Signature · 1 line
2.49.0
#1 lore raw

Hi Nilay,

Quoted 13 lines · Nilay Shroff
drivers/nvme/host/core.c +1 -0
@@ -4089,6 +4089,7 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
 	struct nvme_ns *ns;
 	struct gendisk *disk;
 	int node = ctrl->numa_node;
+	bool last_path = false;
 
 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
 	if (!ns)
@@ -4181,9 +4182,13 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
  out_unlink_ns:
 	mutex_lock(&ctrl->subsys->lock);

drivers/nvme/host/core.c static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info) · line 4192

Quoted 10 lines
drivers/nvme/host/core.c +5 -1
 	list_del_rcu(&ns->siblings);
-	if (list_empty(&ns->head->list))
+	if (list_empty(&ns->head->list)) {
 		list_del_init(&ns->head->entry);
+		last_path = true;
+	}
 	mutex_unlock(&ctrl->subsys->lock);
+	if (last_path)
+		nvme_put_ns_head(ns->head);
 	nvme_put_ns_head(ns->head);

Could the put moved in front of 'goto out_unlink_ns'?

Thanks,
Daniel

#2 lore raw
Quoted 20 lines · Daniel Wagner

Hi Nilay,

Quoted 17 lines · Nilay Shroff
drivers/nvme/host/core.c +2 -1
@@ -4089,6 +4089,7 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
 	struct nvme_ns *ns;
 	struct gendisk *disk;
 	int node = ctrl->numa_node;
+	bool last_path = false;
 
 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
 	if (!ns)
@@ -4181,9 +4182,13 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
  out_unlink_ns:
 	mutex_lock(&ctrl->subsys->lock);
 	list_del_rcu(&ns->siblings);
-	if (list_empty(&ns->head->list))
+	if (list_empty(&ns->head->list)) {
 		list_del_init(&ns->head->entry);
Quoted 8 lines

drivers/nvme/host/core.c static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info) · line 4192

Quoted 6 lines
drivers/nvme/host/core.c +4 -0
+		last_path = true;
+	}
 	mutex_unlock(&ctrl->subsys->lock);
+	if (last_path)
+		nvme_put_ns_head(ns->head);
 	nvme_put_ns_head(ns->head);

Could the put moved in front of 'goto out_unlink_ns'?

I believe you would want to move the second nvme_put_ns_head() above as the first statement under out_unlink_ns label as shown below:

out_unlink_ns:

	nvme_put_ns_head(ns->head);
	mutex_lock(&ctrl->subsys->lock);
	list_del_rcu(&ns->siblings);
	if (list_empty(&ns->head->list)) {
		list_del_init(&ns->head->entry);
                last_path = true;
        }
	mutex_unlock(&ctrl->subsys->lock);
        if (last_path)
                nvme_put_ns_head(ns->head);

If that's the case, and this is indeed the last reference to ns->head, then nvme_put_ns_head() could potentially free the ns->head object. If that happens, any access to ns->head afterward — such as ns->head->list — would result in a use-after-free and potentially trigger a kernel oops, correct?

Thanks,
--Nilay

#3 lore raw
Quoted 18 lines · Nilay Shroff
Quoted 1 line

Could the put moved in front of 'goto out_unlink_ns'?

I believe you would want to move the second nvme_put_ns_head() above as the first statement under out_unlink_ns label as shown below:

out_unlink_ns:

	nvme_put_ns_head(ns->head);
	mutex_lock(&ctrl->subsys->lock);
	list_del_rcu(&ns->siblings);
	if (list_empty(&ns->head->list)) {
		list_del_init(&ns->head->entry);
                last_path = true;
        }
	mutex_unlock(&ctrl->subsys->lock);
        if (last_path)
                nvme_put_ns_head(ns->head);
Quoted 5 lines

If that's the case, and this is indeed the last reference to ns->head, then nvme_put_ns_head() could potentially free the ns->head object. If that happens, any access to ns->head afterward — such as ns->head->list — would result in a use-after-free and potentially trigger a kernel oops, correct?

Ah sorry brainfarth, you are right.

#4 lore raw
In reply to Nilay Shroff
Quoted 1 line · Nilay Shroff

During the failre

failure?

Quoted 3 lines

To fix this, we now unconditionally release the nshead reference when the last nvme path referencing to the nshead is removed, regardless of the head’s state. Also during idnetify

identify

Otherwise this looks good.

#5 lore raw
Quoted 5 lines · Christoph Hellwig
Quoted 1 line · Nilay Shroff

During the failre

failure?

Quoted 7 lines
Quoted 3 lines

To fix this, we now unconditionally release the nshead reference when the last nvme path referencing to the nshead is removed, regardless of the head’s state. Also during idnetify

identify

Otherwise this looks good.

Thanks! I'd update the above typo, however today I was testing some code after disabling CONFIG_NVME_MULTIPATH and I realized that although the namespace head (ns->head) is still created, its associated head->disk is not initialized in this configuration. As a result, ns->head only has a single reference — the one from kref_init() during its creation. Therefore, when we delete the namespace or handle an error case, we must ensure that we only call nvme_put_ns_head() once to properly release the reference and avoid a potential double-free. So I updated this patch as shown below (please note that this is equally applicable when we insmod nvme_core with module param "multipath" set to false):

drivers/nvme/host/core.c +7 -1
index 92697f98c601..033fe0a12727 100644
@@ -4089,6 +4089,7 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
        struct nvme_ns *ns;
        struct gendisk *disk;
        int node = ctrl->numa_node;
+       bool last_path = false;
 
        ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
        if (!ns)
@@ -4181,9 +4182,14 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
  out_unlink_ns:
        mutex_lock(&ctrl->subsys->lock);
        list_del_rcu(&ns->siblings);
-       if (list_empty(&ns->head->list))
+       if (list_empty(&ns->head->list)) {
                list_del_init(&ns->head->entry);
+               if (ns->head->disk)
+                       last_path = true;
+       }
        mutex_unlock(&ctrl->subsys->lock);
+       if (last_path)
+               nvme_put_ns_head(ns->head);
        nvme_put_ns_head(ns->head);
  out_cleanup_disk:
        put_disk(disk);
drivers/nvme/host/multipath.c +4 -1
index e040e467f9fa..c7644631bb1a 100644
@@ -690,8 +690,8 @@ static void nvme_remove_head(struct nvme_ns_head *head)
                nvme_cdev_del(&head->cdev, &head->cdev_device);
                synchronize_srcu(&head->srcu);
                del_gendisk(head->disk);
-               nvme_put_ns_head(head);
        }
+       nvme_put_ns_head(head);
 }
 
 static void nvme_remove_head_work(struct work_struct *work)
@@ -1291,6 +1291,9 @@ void nvme_mpath_remove_disk(struct nvme_ns_head *head)
 {
        bool remove = false;
 
+       if (!head->disk)
+               return;
+
        mutex_lock(&head->subsys->lock);
        /*
         * We are called when all paths have been removed, and at that point

So I'll send out patchv2 with the above changes.

Thanks,
--Nilay

#6 lore raw
Quoted 4 lines · Nilay Shroff

Thanks! I'd update the above typo, however today I was testing some code after disabling CONFIG_NVME_MULTIPATH and I realized that although the namespace head (ns->head) is still created, its associated head->disk is not initialized in this configuration.

Yes.

Quoted 1 line

So I'll send out patchv2 with the above changes.

Thanks! Please also add a comment to nvme_alloc_ns as the logic is fairly complex now and future readers will benefit from an explanation.

#7 lore raw
Quoted 6 lines · Christoph Hellwig
Quoted 4 lines · Nilay Shroff

Thanks! I'd update the above typo, however today I was testing some code after disabling CONFIG_NVME_MULTIPATH and I realized that although the namespace head (ns->head) is still created, its associated head->disk is not initialized in this configuration.

Quoted 8 lines

Yes.

Quoted 1 line

So I'll send out patchv2 with the above changes.

Thanks! Please also add a comment to nvme_alloc_ns as the logic is fairly complex now and future readers will benefit from an explanation.

Yeah will do the needful.

Thanks,
--Nilay