All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH] nvmet: look up the P2P device under rcu_read_lock()
@ 2026-11-16  2:25 Changyul Lee
  2026-11-20 17:43 ` Christoph Hellwig
  0 siblings, 1 reply; 2+ messages in thread
From: Changyul Lee @ 2026-11-16  2:25 UTC (permalink / raw)
  To: Christoph Hellwig, Sagi Grimberg, Chaitanya Kulkarni
  Cc: Logan Gunthorpe, Stephen Bates, Bjorn Helgaas, linux-nvme,
	linux-kernel, Changyul Lee

nvmet_req_find_p2p_dev() looks up ctrl->p2p_ns_map in the I/O path
without subsys->lock and outside an RCU read-side critical section.
nvmet_ns_disable() can meanwhile delete an entry from the same radix
tree, and radix_tree_delete() frees the nodes that become empty with
call_rcu(). With preemptible RCU, the grace period can end while the
lookup still walks such a node, and the lookup reads freed memory.

radix_tree_lookup() without the tree lock must be called under
rcu_read_lock(). Take it around the lookup.

Fixes: c6925093d0b2 ("nvmet: Optionally use PCI P2P memory")
Signed-off-by: Changyul Lee <lcy8047@gmail.com>
---
Found with a static analysis tool we are developing.

Compile-tested only, with x86_64_defconfig plus NVME_TARGET_RDMA,
PCI_P2PDMA and W=1.

 drivers/nvme/target/core.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/drivers/nvme/target/core.c b/drivers/nvme/target/core.c
index e31d65228c11..1b34fc5549f6 100644
--- a/drivers/nvme/target/core.c
+++ b/drivers/nvme/target/core.c
@@ -1341,10 +1341,17 @@ static int nvmet_req_alloc_p2pmem_sgls(struct pci_dev *p2p_dev,
 
 static struct pci_dev *nvmet_req_find_p2p_dev(struct nvmet_req *req)
 {
+	struct pci_dev *p2p_dev;
+
 	if (!IS_ENABLED(CONFIG_PCI_P2PDMA) ||
 	    !req->sq->ctrl || !req->sq->qid || !req->ns)
 		return NULL;
-	return radix_tree_lookup(&req->sq->ctrl->p2p_ns_map, req->ns->nsid);
+
+	rcu_read_lock();
+	p2p_dev = radix_tree_lookup(&req->sq->ctrl->p2p_ns_map, req->ns->nsid);
+	rcu_read_unlock();
+
+	return p2p_dev;
 }
 
 int nvmet_req_alloc_sgls(struct nvmet_req *req)

base-commit: 7475ea8b58616b33c04afb976cf0fbaad9dd63fb
-- 
2.43.0



^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH] nvmet: look up the P2P device under rcu_read_lock()
  2026-11-16  2:25 [PATCH] nvmet: look up the P2P device under rcu_read_lock() Changyul Lee
@ 2026-11-20 17:43 ` Christoph Hellwig
  0 siblings, 0 replies; 2+ messages in thread
From: Christoph Hellwig @ 2026-11-20 17:43 UTC (permalink / raw)
  To: Changyul Lee
  Cc: Sagi Grimberg, Chaitanya Kulkarni, Logan Gunthorpe, Stephen Bates,
	Bjorn Helgaas, linux-nvme, linux-kernel

>  static struct pci_dev *nvmet_req_find_p2p_dev(struct nvmet_req *req)
>  {
> +	struct pci_dev *p2p_dev;
> +
>  	if (!IS_ENABLED(CONFIG_PCI_P2PDMA) ||
>  	    !req->sq->ctrl || !req->sq->qid || !req->ns)
>  		return NULL;
> -	return radix_tree_lookup(&req->sq->ctrl->p2p_ns_map, req->ns->nsid);
> +
> +	rcu_read_lock();
> +	p2p_dev = radix_tree_lookup(&req->sq->ctrl->p2p_ns_map, req->ns->nsid);
> +	rcu_read_unlock();
> +
> +	return p2p_dev;

Not sure what the RCU section is supposed to protect here.  If we
don't get a reference the result can be stale as soon as we're outside
the critical section.  So IFF this is a real bug, it isn't fixed as
easily.



^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-20 17:43 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-16  2:25 [PATCH] nvmet: look up the P2P device under rcu_read_lock() Changyul Lee
2026-11-20 17:43 ` Christoph Hellwig

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.