[PATCH] nvme/host: Fix RCU list traversal to use SRCU primitive

[PATCH] nvme/host: Fix RCU list traversal to use SRCU primitive

[Breno Leitao]

The code currently uses list_for_each_entry_rcu() while holding an SRCU
lock, triggering false positive warnings with CONFIG_PROVE_RCU=y
enabled:

  drivers/nvme/host/core.c:3770 RCU-list traversed in non-reader section!!

While the list is properly protected by SRCU lock, the code uses the wrong
list traversal primitive. Replace list_for_each_entry_rcu() with
list_for_each_entry_srcu() to correctly indicate SRCU-based protection
and eliminate the false warning.

Signed-off-by: Breno Leitao <leitao@debian.org>
Fixes: be647e2c76b2 ("nvme: use srcu for iterating namespace list")
---
Something similar will need to be done for multipath. I will get it done
once I get some feedback about this patch first.
---
 drivers/nvme/host/core.c | 21 ++++++++++++++-------
 1 file changed, 14 insertions(+), 7 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 84cb859a911d09dbe71b2f1ac473ae687c4dc687..3583bae69ef74c6f1fe6d465531a9a09512a6f13 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -3767,7 +3767,8 @@ struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
 	int srcu_idx;
 
 	srcu_idx = srcu_read_lock(&ctrl->srcu);
-	list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
+	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
+				 srcu_read_lock_held(&ctrl->srcu)) {
 		if (ns->head->ns_id == nsid) {
 			if (!nvme_get_ns(ns))
 				continue;
@@ -4851,7 +4852,8 @@ void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
 	int srcu_idx;
 
 	srcu_idx = srcu_read_lock(&ctrl->srcu);
-	list_for_each_entry_rcu(ns, &ctrl->namespaces, list)
+	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
+				 srcu_read_lock_held(&ctrl->srcu))
 		blk_mark_disk_dead(ns->disk);
 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
 }
@@ -4863,7 +4865,8 @@ void nvme_unfreeze(struct nvme_ctrl *ctrl)
 	int srcu_idx;
 
 	srcu_idx = srcu_read_lock(&ctrl->srcu);
-	list_for_each_entry_rcu(ns, &ctrl->namespaces, list)
+	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
+				 srcu_read_lock_held(&ctrl->srcu))
 		blk_mq_unfreeze_queue(ns->queue);
 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
 	clear_bit(NVME_CTRL_FROZEN, &ctrl->flags);
@@ -4876,7 +4879,8 @@ int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
 	int srcu_idx;
 
 	srcu_idx = srcu_read_lock(&ctrl->srcu);
-	list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
+	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
+				 srcu_read_lock_held(&ctrl->srcu)) {
 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
 		if (timeout <= 0)
 			break;
@@ -4892,7 +4896,8 @@ void nvme_wait_freeze(struct nvme_ctrl *ctrl)
 	int srcu_idx;
 
 	srcu_idx = srcu_read_lock(&ctrl->srcu);
-	list_for_each_entry_rcu(ns, &ctrl->namespaces, list)
+	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
+				 srcu_read_lock_held(&ctrl->srcu))
 		blk_mq_freeze_queue_wait(ns->queue);
 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
 }
@@ -4905,7 +4910,8 @@ void nvme_start_freeze(struct nvme_ctrl *ctrl)
 
 	set_bit(NVME_CTRL_FROZEN, &ctrl->flags);
 	srcu_idx = srcu_read_lock(&ctrl->srcu);
-	list_for_each_entry_rcu(ns, &ctrl->namespaces, list)
+	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
+				 srcu_read_lock_held(&ctrl->srcu))
 		blk_freeze_queue_start(ns->queue);
 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
 }
@@ -4953,7 +4959,8 @@ void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
 	int srcu_idx;
 
 	srcu_idx = srcu_read_lock(&ctrl->srcu);
-	list_for_each_entry_rcu(ns, &ctrl->namespaces, list)
+	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
+				 srcu_read_lock_held(&ctrl->srcu))
 		blk_sync_queue(ns->queue);
 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
 }

---
base-commit: f488649e40f8900d23b86afeab7d4b78c063d5d1
change-id: 20241104-nvme_rcu-8250c60278e6

Best regards,
-- 
Breno Leitao <leitao@debian.org>

Re: [PATCH] nvme/host: Fix RCU list traversal to use SRCU primitive

[Christoph Hellwig]

On Mon, Nov 04, 2024 at 04:24:40AM -0800, Breno Leitao wrote:
> The code currently uses list_for_each_entry_rcu() while holding an SRCU
> lock, triggering false positive warnings with CONFIG_PROVE_RCU=y
> enabled:
> 
>   drivers/nvme/host/core.c:3770 RCU-list traversed in non-reader section!!
> 
> While the list is properly protected by SRCU lock, the code uses the wrong
> list traversal primitive. Replace list_for_each_entry_rcu() with
> list_for_each_entry_srcu() to correctly indicate SRCU-based protection
> and eliminate the false warning.

I didn't even know there was such as thing as list_for_each_entry_srcu,
but apparently it's been there for a while.  Looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>

> Something similar will need to be done for multipath. I will get it done
> once I get some feedback about this patch first.

Thanks!

Re: [PATCH] nvme/host: Fix RCU list traversal to use SRCU primitive

[Keith Busch]

On Mon, Nov 04, 2024 at 02:16:52PM +0100, Christoph Hellwig wrote:
> On Mon, Nov 04, 2024 at 04:24:40AM -0800, Breno Leitao wrote:
> > The code currently uses list_for_each_entry_rcu() while holding an SRCU
> > lock, triggering false positive warnings with CONFIG_PROVE_RCU=y
> > enabled:
> > 
> >   drivers/nvme/host/core.c:3770 RCU-list traversed in non-reader section!!
> > 
> > While the list is properly protected by SRCU lock, the code uses the wrong
> > list traversal primitive. Replace list_for_each_entry_rcu() with
> > list_for_each_entry_srcu() to correctly indicate SRCU-based protection
> > and eliminate the false warning.
> 
> I didn't even know there was such as thing as list_for_each_entry_srcu,
> but apparently it's been there for a while.  Looks good:

Neither did I! Thanks Breno, applied to nvme-6.12