[PATCH 01/10] nvmet: zeroout id-ns buffer for invalid nsid

Christoph Hellwig hch at lst.de
Mon Feb 1 12:24:38 EST 2021


On Sun, Jan 31, 2021 at 09:41:29PM -0800, Chaitanya Kulkarni wrote:
> According to spec :-
> "5.15.2.7 Identify Namespace data structure for an Allocated Namespace
> ID (CNS 11h) The Identify Namespace data structure (refer to Figure 245)
> is returned to the host for the namespace specified in the Namespace
> Identifier (NSID) field if it is an allocated NSID. If the specified
> namespace is an unallocated NSID then the controller returns a zero
> filled data structure."
> 
> Move call to nvmet_find_namespace() in nvmet_execute_identify_ns()
> above the id-ns buffer allocation since there is no point in allocating
> a buffer if namespace doesn't exist. Call nvmet_zero_sgl() when call to
> nvmet_find_namespace() fails and add an explicit comment to specify the
> reason for zeroing the buffer which is not common for the NVMe commands.
> Remove the done label and we can directly jump to out label from
> nvmet_find_namespace() error case.
> 
> Fixes: bffcd507780e ("nvmet: set right status on error in id-ns handler")
> Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni at wdc.com>
> ---
>  drivers/nvme/target/admin-cmd.c | 22 ++++++++++++++--------
>  1 file changed, 14 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/nvme/target/admin-cmd.c b/drivers/nvme/target/admin-cmd.c
> index 613a4d8feac1..8cc7bb25d10d 100644
> --- a/drivers/nvme/target/admin-cmd.c
> +++ b/drivers/nvme/target/admin-cmd.c
> @@ -476,19 +476,25 @@ static void nvmet_execute_identify_ns(struct nvmet_req *req)
>  		goto out;
>  	}
>  
> +	req->ns = nvmet_find_namespace(ctrl, req->cmd->identify.nsid);
> +	if (!req->ns) {
> +		status = NVME_SC_INVALID_NS;
> +		/*
> +		 * According to spec : If the specified namespace is
> +		 * an unallocated NSID then the controller returns a zero filled
> +		 * data structure. Also don't override the error status as invalid
> +		 * namespace takes priority over the failed zeroout buffer case.
> +		 */
> +		nvmet_zero_sgl(req, 0, sizeof(*id));
> +		goto out;
> +	}
> +
>  	id = kzalloc(sizeof(*id), GFP_KERNEL);
>  	if (!id) {
>  		status = NVME_SC_INTERNAL;
>  		goto out;
>  	}
>  
> -	/* return an all zeroed buffer if we can't find an active namespace */
> -	req->ns = nvmet_find_namespace(ctrl, req->cmd->identify.nsid);
> -	if (!req->ns) {
> -		status = NVME_SC_INVALID_NS;
> -		goto done;

I think all we need to do is to remove the status assignment here.
While you're patch avoids the memory allocation for this case, it isn't
really the fast path so I'd rather avoid the extra code.



More information about the Linux-nvme mailing list