[PATCH V2 3/3] nvmet: add ns write protect support

[hch@infradead.org (Christoph Hellwig)]

On Thu, Jul 26, 2018@07:56:56PM -0700, Chaitanya Kulkarni wrote:
> This patch implements the Namespace Write Protect feature described in
> "NVMe TP 4005a Namespace Write Protect". In this version, we implement
> No Write Protect and Write Protect states for target ns which can be
> toggled by set-features commands from the host side.
> 
> For write-protect state transition, we need to flush the ns specified
> as a part of command so we also add helpers for carrying out synchronous
> flush operations.
> 
> Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni at wdc.com>
> ---
>  drivers/nvme/target/admin-cmd.c   | 120 +++++++++++++++++++++++++++++-
>  drivers/nvme/target/core.c        |   6 +-
>  drivers/nvme/target/io-cmd-bdev.c |  12 +++
>  drivers/nvme/target/io-cmd-file.c |  15 +++-
>  drivers/nvme/target/nvmet.h       |   6 ++
>  include/linux/nvme.h              |   2 +-
>  6 files changed, 154 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/nvme/target/admin-cmd.c b/drivers/nvme/target/admin-cmd.c
> index 16a9b24270f9..17fb512d5c21 100644
> --- a/drivers/nvme/target/admin-cmd.c
> +++ b/drivers/nvme/target/admin-cmd.c
> @@ -19,6 +19,51 @@
>  #include <asm/unaligned.h>
>  #include "nvmet.h"
>  
> +bool nvmet_ns_wp_cmd_allow(struct nvmet_req *req)
> +{
> +	bool ret = false;
> +
> +	if (likely(req->sq->qid != 0)) {
> +		switch (req->cmd->common.opcode) {
> +		case nvme_cmd_read:
> +		case nvme_cmd_flush:
> +			/* fall thru */
> +			ret = true;
> +		}
> +		return ret;
> +	}
> +	/*
> +	 * Right now we don't have a useful way to check for the allowed admin
> +	 * cmds when the ns is write-protected, as we don't have any admin-cmds
> +	 * which are operating on the target ns except for write-protect-feat.
> +	 * For the completeness keep the valid list of admin cmds from the spec
> +	 * here. In future when we add a new cmd or implement a feature
> +	 * which operates on a ns and will trigger media change please add this
> +	 * call to the code in the appropriate location and update cmd list.
> +	 */

I guess we could short cut most of this by checking for a nsid of 0
or 0xffffffff first and always allow, otherwise reject.

> +static u16 nvmet_write_protect_flush_sync(struct nvmet_req *req)
> +{
> +	u16 status;
> +
> +	status = req->ns->file ? nvmet_file_flush(req) : nvmet_bdev_flush(req);

Nit: I think this would be more readable as a classic if / else:

	if (req->ns->file)
		status = nvmet_file_flush(req);
	else
		status = nvmet_bdev_flush(req);
		

> +
> +	if (status) {
> +		pr_err("write protect flush failed nsid: %u\n", req->ns->nsid);
> +		status = NVME_SC_INTERNAL | NVME_SC_DNR;
> +	}
> +	return status;

Why do we discard the original status here?

> +static u16 nvmet_feat_write_protect(struct nvmet_req *req)
> +{
> +	u32 write_protect = le32_to_cpu(req->cmd->common.cdw10[1]);
> +	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
> +	u16 status = NVME_SC_FEATURE_NOT_CHANGEABLE;
> +
> +	req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->rw.nsid);
> +	if (unlikely(!req->ns))
> +		return status;
> +
> +	if (req->ns->readonly == false) {
> +		if (write_protect == NVME_NS_WRITE_PROTECT) {
> +			req->ns->readonly = true;

We need some synchronization around the whole setting of this flag.

> +			status = nvmet_write_protect_flush_sync(req);

I think we need to undo the readonly flag and exit early if we get
a non-zero status here.

> +			/* Generate async event */
> +			mutex_lock(&subsys->lock);
> +			nvmet_ns_changed(subsys, req->ns->nsid);
> +			mutex_unlock(&subsys->lock);

I'd rather keep this code in a single place at the end of the function
instead of duplicating it.  Something like:

	mutex_lock(&subsys->lock);
	switch (write_protect) {
	case NVME_NS_WRITE_PROTECT:
		req->ns->readonly = true;
		status = nvmet_write_protect_flush_sync(req);
		if (status)
			req->ns->readonly = false;
		break;
	case NVME_NS_NO_WRITE_PROTECT:
		req->ns->readonly = false;
		status = NVME_SC_SUCCESS;
		break;
	default:
		break;
	}

	if (!status)
		nvmet_ns_changed(subsys, req->ns->nsid);
	mutex_unlock(&subsys->lock);

	return status;

>  		break;
> +	case NVME_FEAT_WRITE_PROTECT:
> +		req->ns = nvmet_find_namespace(req->sq->ctrl, nsid);
> +		if (!req->ns)
> +			status = NVME_SC_INVALID_NS | NVME_SC_DNR;
> +		else {
> +			if (req->ns->readonly == true)
> +				result = NVME_NS_WRITE_PROTECT;
> +			else
> +				result = NVME_NS_NO_WRITE_PROTECT;
> +			nvmet_set_result(req, result);
> +		}

Probably nicer to split this into a little helper function.

>  {
>  	struct nvmet_ctrl *ctrl;
>  
> @@ -443,6 +443,7 @@ struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid)
>  	ns->subsys = subsys;
>  	uuid_gen(&ns->uuid);
>  	ns->buffered_io = false;
> +	ns->readonly = false;

ns is already allocated using kzalloc, so this isn't needed.

>  
>  	return ns;
>  }
> @@ -561,6 +562,9 @@ static u16 nvmet_parse_io_cmd(struct nvmet_req *req)
>  	if (unlikely(!req->ns))
>  		return NVME_SC_INVALID_NS | NVME_SC_DNR;
>  
> +	if (unlikely(req->ns->readonly && nvmet_ns_wp_cmd_allow(req) == false))
> +		return NVME_SC_NS_WRITE_PROTECTED;

I'd simplify this to:

	if (unlikely(req->ns->readonly && !nvmet_ns_wp_cmd_allow(req))
		return NVME_SC_NS_WRITE_PROTECTED;

> +
>  	if (req->ns->file)
>  		return nvmet_file_parse_io_cmd(req);
>  	else
> diff --git a/drivers/nvme/target/io-cmd-bdev.c b/drivers/nvme/target/io-cmd-bdev.c
> index e0b0f7df70c2..e6819ad8d153 100644
> --- a/drivers/nvme/target/io-cmd-bdev.c
> +++ b/drivers/nvme/target/io-cmd-bdev.c
> @@ -124,6 +124,18 @@ static void nvmet_bdev_execute_flush(struct nvmet_req *req)
>  	submit_bio(bio);
>  }
>  
> +u16 nvmet_bdev_flush(struct nvmet_req *req)
> +{
> +	blk_status_t status;
> +	int ret;
> +
> +	ret = blkdev_issue_flush(req->ns->bdev, GFP_KERNEL, NULL);
> +
> +	status = errno_to_blk_status(ret);
> +
> +	return status != BLK_STS_OK ? NVME_SC_INTERNAL | NVME_SC_DNR : 0;

Hmm, why not:

	ret = blkdev_issue_flush(req->ns->bdev, GFP_KERNEL, NULL);

	return ret ? NVME_SC_INTERNAL | NVME_SC_DNR : 0;

> +}
> +
>  static u16 nvmet_bdev_discard_range(struct nvmet_ns *ns,
>  		struct nvme_dsm_range *range, struct bio **bio)
>  {
> diff --git a/drivers/nvme/target/io-cmd-file.c b/drivers/nvme/target/io-cmd-file.c
> index c2d0d08b59c8..152b2ef236e1 100644
> --- a/drivers/nvme/target/io-cmd-file.c
> +++ b/drivers/nvme/target/io-cmd-file.c
> @@ -211,14 +211,23 @@ static void nvmet_file_execute_rw_buffered_io(struct nvmet_req *req)
>  	queue_work(buffered_io_wq, &req->f.work);
>  }
>  
> -static void nvmet_file_flush_work(struct work_struct *w)
> +u16 nvmet_file_flush(struct nvmet_req *req)
>  {
> -	struct nvmet_req *req = container_of(w, struct nvmet_req, f.work);
>  	int ret;
>  
>  	ret = vfs_fsync(req->ns->file, 1);
>  
> -	nvmet_req_complete(req, ret < 0 ? NVME_SC_INTERNAL | NVME_SC_DNR : 0);
> +	return ret < 0 ? NVME_SC_INTERNAL | NVME_SC_DNR : 0;
> +}
> +
> +static void nvmet_file_flush_work(struct work_struct *w)
> +{
> +	struct nvmet_req *req = container_of(w, struct nvmet_req, f.work);
> +	u16 status;
> +
> +	status = nvmet_file_flush(req);
> +
> +	nvmet_req_complete(req, status);
>  }
>  
>  static void nvmet_file_execute_flush(struct nvmet_req *req)
> diff --git a/drivers/nvme/target/nvmet.h b/drivers/nvme/target/nvmet.h
> index 688993855402..d9171faaa26d 100644
> --- a/drivers/nvme/target/nvmet.h
> +++ b/drivers/nvme/target/nvmet.h
> @@ -60,6 +60,7 @@ struct nvmet_ns {
>  	struct block_device	*bdev;
>  	struct file		*file;
>  	u32			nsid;
> +	bool			readonly;
>  	u32			blksize_shift;
>  	loff_t			size;
>  	u8			nguid[16];
> @@ -380,6 +381,11 @@ int nvmet_bdev_ns_enable(struct nvmet_ns *ns);
>  int nvmet_file_ns_enable(struct nvmet_ns *ns);
>  void nvmet_bdev_ns_disable(struct nvmet_ns *ns);
>  void nvmet_file_ns_disable(struct nvmet_ns *ns);
> +u16 nvmet_bdev_flush(struct nvmet_req *req);
> +u16 nvmet_file_flush(struct nvmet_req *req);
> +int nvmet_ns_is_wp(struct nvmet_ns *ns);
> +bool nvmet_ns_wp_cmd_allow(struct nvmet_req *req);
> +void nvmet_ns_changed(struct nvmet_subsys *subsys, u32 nsid);
>  
>  static inline u32 nvmet_rw_len(struct nvmet_req *req)
>  {
> diff --git a/include/linux/nvme.h b/include/linux/nvme.h
> index 8514d4e0b597..d02561cf94c8 100644
> --- a/include/linux/nvme.h
> +++ b/include/linux/nvme.h
> @@ -312,7 +312,7 @@ struct nvme_id_ns {
>  	__le16			nabspf;
>  	__le16			noiob;
>  	__u8			nvmcap[16];
> -	__u8			rsvd64[32];
> +	__u8			rsvd64[35];

Ok, looks like this needs to be folded into the first patch.