On May 21, 2025 / 20:51, Shin'ichiro Kawasaki wrote:
[...]
> With this fix trial patch, the KASAN sauf is still observed. I guess it has
> another cause and requires more debug work.
I chased down the KASAN suaf, and now I think I understand the cause. When it
happens, nvme_tcp_create_ctrl() fails to create the nvme- tcp control in the
call chain below:
nvme_tcp_create_ctrl()
nvme_tcp_alloc_ctrl() new=true ... Alloc nvme_tcp_ctrl and admin_tag_set
nvme_tcp_setup_ctrl() new=true
nvme_tcp_configure_admin_queue() new=true ... Succeed
nvme_alloc_admin_tag_set() ... Alloc the tag set for admin_tag_set
nvme_stop_keep_alive()
nvme_tcp_teardown_admin_queue() remove=false
nvme_tcp_configure_admin_queue() new=false
nvme_tcp_alloc_admin_queue() ... Fail, but do not call nvme_remove_admin_tag_set()
nvme_uninit_ctrl()
nvme_put_ctrl() ... Free up the nvme_tcp_ctrl and admin_tag_set
In this call chain, the first call of nvme_tcp_configure_admin_queue()
succeeds with new=true argument. The second call fails with new=false
argument. This second call does not call nvme_remove_admin_tag_set(),
due to the new=false argument. Then the admin tag set is not removed.
nvme_tcp_create_ctrl() assumes that nvme_tcp_setup_ctrl() would call
nvme_remove_admin_tag_set(), and frees up struct nvme_tcp_ctrl which has
admin_tag_set field. Later on, the timeout handler accesses the
admin_tag_set field and causes the BUG KASAN slab-use-after-free.
I created a trial patch below. When the second
nvme_tcp_configure_admin_queue() call fails, it jumps to "destroy_admin"
go-to label to call nvme_tcp_teardown_admin_queue() which calls
nvme_remove_admin_tag_set(). With this fix, the KASAN suaf looks disappearing.
I will create a formal patch for review. I will post it as a series, which will
have two patches: one for this KASAN suaf, the other for the WARN.
diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
index d89c89570d11..74a388550995 100644
--- a/drivers/nvme/host/tcp.c
+++ b/drivers/nvme/host/tcp.c
@@ -2392,7 +2392,7 @@ static int nvme_tcp_setup_ctrl(struct nvme_ctrl *ctrl, bool new)
nvme_tcp_teardown_admin_queue(ctrl, false);
ret = nvme_tcp_configure_admin_queue(ctrl, false);
if (ret)
- return ret;
+ goto destroy_admin;
}
if (ctrl->icdoff) {
