lkml.org 
[lkml]   [2018]   [May]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH] nvme-rdma: clear NVME_RDMA_Q_LIVE before free the queue
From
Date


On 05/04/2018 11:02 AM, Jianchao Wang wrote:
> When nvme_init_identify in nvme_rdma_configure_admin_queue fails,
> the ctrl->queues[0] is freed but the NVME_RDMA_Q_LIVE is still set.
> If nvme_rdma_stop_queue is invoked, we will incur use-after-free
> which will cause memory corruption.
> BUG: KASAN: use-after-free in rdma_disconnect+0x1f/0xe0 [rdma_cm]
> Read of size 8 at addr ffff8801dc3969c0 by task kworker/u16:3/9304
>
> CPU: 3 PID: 9304 Comm: kworker/u16:3 Kdump: loaded Tainted: G W 4.17.0-rc3+ #20
> Workqueue: nvme-delete-wq nvme_delete_ctrl_work
> Call Trace:
> dump_stack+0x91/0xeb
> print_address_description+0x6b/0x290
> kasan_report+0x261/0x360
> rdma_disconnect+0x1f/0xe0 [rdma_cm]
> nvme_rdma_stop_queue+0x25/0x40 [nvme_rdma]
> nvme_rdma_shutdown_ctrl+0xf3/0x150 [nvme_rdma]
> nvme_delete_ctrl_work+0x98/0xe0
> process_one_work+0x3ca/0xaa0
> worker_thread+0x4e2/0x6c0
> kthread+0x18d/0x1e0
> ret_from_fork+0x24/0x30
>
> To fix it, clear the NVME_RDMA_Q_LIVE before free the ctrl->queues[0].
> The queue will be freed, so it certainly is not LIVE any more.
>
> Signed-off-by: Jianchao Wang <jianchao.w.wang@oracle.com>
> ---
> drivers/nvme/host/rdma.c | 5 +++++
> 1 file changed, 5 insertions(+)
>
> diff --git a/drivers/nvme/host/rdma.c b/drivers/nvme/host/rdma.c
> index fd965d0..ffbfe82 100644
> --- a/drivers/nvme/host/rdma.c
> +++ b/drivers/nvme/host/rdma.c
> @@ -812,6 +812,11 @@ static int nvme_rdma_configure_admin_queue(struct nvme_rdma_ctrl *ctrl,
> if (new)
> nvme_rdma_free_tagset(&ctrl->ctrl, ctrl->ctrl.admin_tagset);
> out_free_queue:
> + /*
> + * The queue will be freed, so it is not LIVE any more.
> + * This could avoid use-after-free in nvme_rdma_stop_queue.
> + */
> + clear_bit(NVME_RDMA_Q_LIVE, &ctrl->queues[0].flags);
> nvme_rdma_free_queue(&ctrl->queues[0]);
> return error;
> }
>

The correct fix would be to add a tag for stop_queue and call
nvme_rdma_stop_queue() in all the failure cases after
nvme_rdma_start_queue.

\
 
 \ /
  Last update: 2018-05-09 17:07    [W:0.095 / U:0.308 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site