Re: [PATCHv3 2/2] nvme-fabrics: fix kernel crash while shutting down controller

[Ming Lei <ming.lei@redhat.com>]

On Sun, Nov 03, 2024 at 11:44:42PM +0530, Nilay Shroff wrote:
> The nvme keep-alive operation, which executes at a periodic interval,
> could potentially sneak in while shutting down a fabric controller.
> This may lead to a race between the fabric controller admin queue
> destroy code path (invoked while shutting down controller) and hw/hctx
> queue dispatcher called from the nvme keep-alive async request queuing
> operation. This race could lead to the kernel crash shown below:
> 
> Call Trace:
>     autoremove_wake_function+0x0/0xbc (unreliable)
>     __blk_mq_sched_dispatch_requests+0x114/0x24c
>     blk_mq_sched_dispatch_requests+0x44/0x84
>     blk_mq_run_hw_queue+0x140/0x220
>     nvme_keep_alive_work+0xc8/0x19c [nvme_core]
>     process_one_work+0x200/0x4e0
>     worker_thread+0x340/0x504
>     kthread+0x138/0x140
>     start_kernel_thread+0x14/0x18
> 
> While shutting down fabric controller, if nvme keep-alive request sneaks
> in then it would be flushed off. The nvme_keep_alive_end_io function is
> then invoked to handle the end of the keep-alive operation which
> decrements the admin->q_usage_counter and assuming this is the last/only
> request in the admin queue then the admin->q_usage_counter becomes zero.
> If that happens then blk-mq destroy queue operation (blk_mq_destroy_
> queue()) which could be potentially running simultaneously on another
> cpu (as this is the controller shutdown code path) would forward
> progress and deletes the admin queue. So, now from this point onward
> we are not supposed to access the admin queue resources. However the
> issue here's that the nvme keep-alive thread running hw/hctx queue
> dispatch operation hasn't yet finished its work and so it could still
> potentially access the admin queue resource while the admin queue had
> been already deleted and that causes the above crash.
> 
> The above kernel crash is regression caused due to changes implemented
> in commit a54a93d0e359 ("nvme: move stopping keep-alive into
> nvme_uninit_ctrl()"). Ideally we should stop keep-alive before destroyin
> g the admin queue and freeing the admin tagset so that it wouldn't sneak
> in during the shutdown operation. However we removed the keep alive stop
> operation from the beginning of the controller shutdown code path in commit
> a54a93d0e359 ("nvme: move stopping keep-alive into nvme_uninit_ctrl()")
> and added it under nvme_uninit_ctrl() which executes very late in the
> shutdown code path after the admin queue is destroyed and its tagset is
> removed. So this change created the possibility of keep-alive sneaking in
> and interfering with the shutdown operation and causing observed kernel
> crash.
> 
> To fix the observed crash, we decided to move nvme_stop_keep_alive() from
> nvme_uninit_ctrl() to nvme_remove_admin_tag_set(). This change would ensure
> that we don't forward progress and delete the admin queue until the keep-
> alive operation is finished (if it's in-flight) or cancelled and that would
> help contain the race condition explained above and hence avoid the crash.
> 
> Fixes: a54a93d0e359 ("nvme: move stopping keep-alive into nvme_uninit_ctrl()")
> Link: https://lore.kernel.org/all/1a21f37b-0f2a-4745-8c56-4dc8628d3983@linux.ibm.com/
> Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
> ---
>  drivers/nvme/host/core.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
> index ddf07df243d3..a55c56123bfe 100644
> --- a/drivers/nvme/host/core.c
> +++ b/drivers/nvme/host/core.c
> @@ -4593,6 +4593,11 @@ EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
>  
>  void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
>  {
> +	/*
> +	 * As we're about to destroy the queue and free tagset
> +	 * we can not have keep-alive work running.
> +	 */
> +	nvme_stop_keep_alive(ctrl);
>  	blk_mq_destroy_queue(ctrl->admin_q);
>  	blk_put_queue(ctrl->admin_q);
>  	if (ctrl->ops->flags & NVME_F_FABRICS) {

Looks fine,

Reviewed-by: Ming Lei <ming.lei@redhat.com>

Thanks,
Ming