From: Peijie Shao <shaopeijie@cestc.cn>

In a SELinux enabled kernel, socket_create() initializes the
security label of the socket using the security label of the
calling process, this typically works well.

However, in a containerized environment like Kubernetes,
problem arises when a privileged container(domain spc_t)
connects to an NVMe target and mounts the NVMe as persistent
storage for unprivileged containers(domain container_t).

This is because the container_t domain cannot access
resources labeled with spc_t, resulting in socket_sendmsg
returning -EACCES.

The solution is to use socket_create_kern() instead of
socket_create(), which labels the socket context to kernel_t.
Access control will then be handled by the VFS layer rather
than the socket itself.

Signed-off-by: Peijie Shao <shaopeijie@cestc.cn>
---
 drivers/nvme/host/tcp.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
index 327f3f2f5399..71f7045881eb 100644
--- a/drivers/nvme/host/tcp.c
+++ b/drivers/nvme/host/tcp.c
@@ -1790,7 +1790,8 @@ static int nvme_tcp_alloc_queue(struct nvme_ctrl *nctrl, int qid,
 		queue->cmnd_capsule_len = sizeof(struct nvme_command) +
 						NVME_TCP_ADMIN_CCSZ;
 
-	ret = sock_create(ctrl->addr.ss_family, SOCK_STREAM,
+	ret = sock_create_kern(current->nsproxy->net_ns,
+			ctrl->addr.ss_family, SOCK_STREAM,
 			IPPROTO_TCP, &queue->sock);
 	if (ret) {
 		dev_err(nctrl->device,
-- 
2.43.0

On Thu, Mar 20, 2025 at 02:35:23PM +0800, shaopeijie@cestc.cn wrote:
> From: Peijie Shao <shaopeijie@cestc.cn>
> 
> In a SELinux enabled kernel, socket_create() initializes the
> security label of the socket using the security label of the
> calling process, this typically works well.
> 
> However, in a containerized environment like Kubernetes,
> problem arises when a privileged container(domain spc_t)
> connects to an NVMe target and mounts the NVMe as persistent
> storage for unprivileged containers(domain container_t).
> 
> This is because the container_t domain cannot access
> resources labeled with spc_t, resulting in socket_sendmsg
> returning -EACCES.
> 
> The solution is to use socket_create_kern() instead of
> socket_create(), which labels the socket context to kernel_t.
> Access control will then be handled by the VFS layer rather
> than the socket itself.

Thanks, applied to nvme-6.15.

Looks good:

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

Can you send a patch for the target side as well?