[PATCH] nvmet: pci-epf: Do not complete commands twice if nvmet_req_init() fails

[Damien Le Moal]

On 7/15/25 18:18, Rick Wertenbroek wrote:
> Have nvmet_req_init() and req->execute() complete failed commands.
> 
> Description of the problem:
> nvmet_req_init() calls __nvmet_req_complete() internally upon failure,
> e.g., unsupported opcode, which calls the "queue_response" callback,
> this results in nvmet_pci_epf_queue_response() being called, which will
> call nvmet_pci_epf_complete_iod() if data_len is 0 or if dma_dir is
> different than DMA_TO_DEVICE. This results in a double completion as
> nvmet_pci_epf_exec_iod_work() also calls nvmet_pci_epf_complete_iod()
> when nvmet_req_init() fails.
> 
> Steps to reproduce:
> On the host send a command with an unsupported opcode with nvme-cli,
> For example the admin command "security receive"
> $ sudo nvme security-recv /dev/nvme0n1 -n1 -x4096
> 
> This triggers a double completion as nvmet_req_init() fails and
> nvmet_pci_epf_queue_response() is called, here iod->dma_dir is still
> in the default state of "DMA_NONE" as set by default in
> nvmet_pci_epf_alloc_iod(), so nvmet_pci_epf_complete_iod() is called.
> Because nvmet_req_init() failed nvmet_pci_epf_complete_iod() is also
> called in nvmet_pci_epf_exec_iod_work() leading to a doubple completion.
> 
> This patch lets nvmet_req_init() and req->execute() complete all failed
> commands, and removes the double completion case in
> nvmet_pci_epf_exec_iod_work() therefore fixing the edge cases where
> double completions occurred.
> 
> Signed-off-by: Rick Wertenbroek <rick.wertenbroek at gmail.com>
> ---
>  drivers/nvme/target/pci-epf.c | 18 ++++++++++++------
>  1 file changed, 12 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/nvme/target/pci-epf.c b/drivers/nvme/target/pci-epf.c
> index a4295a5b8d28..aad828eb72d6 100644
> --- a/drivers/nvme/target/pci-epf.c
> +++ b/drivers/nvme/target/pci-epf.c
> @@ -1243,7 +1243,7 @@ static void nvmet_pci_epf_queue_response(struct nvmet_req *req)
>  	iod->status = le16_to_cpu(req->cqe->status) >> 1;
>  
>  	/* If we have no data to transfer, directly complete the command. */

Maybe change this comment to:

	/*
	 * If the command failed or we have no data to transfer, complete
	 * the command immediately.
	 */

Other than this, this looks good, so feel free to add:

Reviewed-by: Damien Le Moal <dlemoal at kernel.org>

> -	if (!iod->data_len || iod->dma_dir != DMA_TO_DEVICE) {
> +	if (iod->status || !iod->data_len || iod->dma_dir != DMA_TO_DEVICE) {
>  		nvmet_pci_epf_complete_iod(iod);
>  		return;
>  	}
> @@ -1604,8 +1604,13 @@ static void nvmet_pci_epf_exec_iod_work(struct work_struct *work)
>  		goto complete;
>  	}
>  
> +	/*
> +	 * If nvmet_req_init() fails (e.g., unsupported opcode) it will call
> +	 * __nvmet_req_complete() internally which will call
> +	 * nvmet_pci_epf_queue_response() and will complete the command directly.
> +	 */
>  	if (!nvmet_req_init(req, &iod->sq->nvme_sq, &nvmet_pci_epf_fabrics_ops))
> -		goto complete;
> +		return;
>  
>  	iod->data_len = nvmet_req_transfer_len(req);
>  	if (iod->data_len) {
> @@ -1643,10 +1648,11 @@ static void nvmet_pci_epf_exec_iod_work(struct work_struct *work)
>  
>  	wait_for_completion(&iod->done);
>  
> -	if (iod->status == NVME_SC_SUCCESS) {
> -		WARN_ON_ONCE(!iod->data_len || iod->dma_dir != DMA_TO_DEVICE);
> -		nvmet_pci_epf_transfer_iod_data(iod);
> -	}
> +	if (iod->status != NVME_SC_SUCCESS)
> +		return;
> +
> +	WARN_ON_ONCE(!iod->data_len || iod->dma_dir != DMA_TO_DEVICE);
> +	nvmet_pci_epf_transfer_iod_data(iod);
>  
>  complete:
>  	nvmet_pci_epf_complete_iod(iod);


-- 
Damien Le Moal
Western Digital Research