diff mbox series

nvmet: fix the use of ZERO_PAGE in nvme_execute_identify_ns_nvm()

Message ID 20241122085113.2487839-1-nilay@linux.ibm.com (mailing list archive)
State New
Headers show
Series nvmet: fix the use of ZERO_PAGE in nvme_execute_identify_ns_nvm() | expand

Commit Message

Nilay Shroff Nov. 22, 2024, 8:50 a.m. UTC
The nvme_execute_identify_ns_nvm function uses ZERO_PAGE
for copying SG list with all zeros. As ZERO_PAGE would not
necessarily return the virtual-address of the zero page, we
need to first convert the page address to kernel virtual-
address and then use it as source address for copying the
data to SG list with all zeros.

Using return address of ZERO_PAGE(0) as source address for
copying data to SG list would fill the target buffer with
random value and causes the undesired side effect. This patch
implements the fix ensuring that we use virtual-address of the
zero page for copying all zeros to the SG list buffers.

Link: https://lore.kernel.org/all/CAHj4cs8OVyxmn4XTvA=y4uQ3qWpdw-x3M3FSUYr-KpE-nhaFEA@mail.gmail.com/
Fixes: 64a51080eaba ("nvmet: implement id ns for nvm command set")
[nilay: Use page_to_virt() for converting ZERO_PAGE address to
        virtual-address as suggested by Maurizio Lombardi]
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
---
 drivers/nvme/target/admin-cmd.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

Comments

Nilay Shroff Nov. 22, 2024, 8:57 a.m. UTC | #1
Sorry but I forgot to add the reported-by tag.

Reported-by: Yi Zhang <yi.zhang@redhat.com>

On 11/22/24 14:20, Nilay Shroff wrote:
> The nvme_execute_identify_ns_nvm function uses ZERO_PAGE
> for copying SG list with all zeros. As ZERO_PAGE would not
> necessarily return the virtual-address of the zero page, we
> need to first convert the page address to kernel virtual-
> address and then use it as source address for copying the
> data to SG list with all zeros.
> 
> Using return address of ZERO_PAGE(0) as source address for
> copying data to SG list would fill the target buffer with
> random value and causes the undesired side effect. This patch
> implements the fix ensuring that we use virtual-address of the
> zero page for copying all zeros to the SG list buffers.
> 
> Link: https://lore.kernel.org/all/CAHj4cs8OVyxmn4XTvA=y4uQ3qWpdw-x3M3FSUYr-KpE-nhaFEA@mail.gmail.com/
> Fixes: 64a51080eaba ("nvmet: implement id ns for nvm command set")
> [nilay: Use page_to_virt() for converting ZERO_PAGE address to
>         virtual-address as suggested by Maurizio Lombardi]
> Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
> ---
>  drivers/nvme/target/admin-cmd.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/nvme/target/admin-cmd.c b/drivers/nvme/target/admin-cmd.c
> index 934b401fbc2f..a2b0444f28ab 100644
> --- a/drivers/nvme/target/admin-cmd.c
> +++ b/drivers/nvme/target/admin-cmd.c
> @@ -901,12 +901,14 @@ static void nvmet_execute_identify_ctrl_nvm(struct nvmet_req *req)
>  static void nvme_execute_identify_ns_nvm(struct nvmet_req *req)
>  {
>  	u16 status;
> +	void *zero_buf;
>  
>  	status = nvmet_req_find_ns(req);
>  	if (status)
>  		goto out;
>  
> -	status = nvmet_copy_to_sgl(req, 0, ZERO_PAGE(0),
> +	zero_buf = page_to_virt(ZERO_PAGE(0));
> +	status = nvmet_copy_to_sgl(req, 0, zero_buf,
>  				   NVME_IDENTIFY_DATA_SIZE);
>  out:
>  	nvmet_req_complete(req, status);
Maurizio Lombardi Nov. 22, 2024, 9:29 a.m. UTC | #2
pá 22. 11. 2024 v 9:51 odesílatel Nilay Shroff <nilay@linux.ibm.com> napsal:
>  static void nvme_execute_identify_ns_nvm(struct nvmet_req *req)
>  {
>         u16 status;
> +       void *zero_buf;
>
>         status = nvmet_req_find_ns(req);
>         if (status)
>                 goto out;
>
> -       status = nvmet_copy_to_sgl(req, 0, ZERO_PAGE(0),
> +       zero_buf = page_to_virt(ZERO_PAGE(0));
> +       status = nvmet_copy_to_sgl(req, 0, zero_buf,
>                                    NVME_IDENTIFY_DATA_SIZE);
>  out:
>         nvmet_req_complete(req, status);

I will later submit a patch to ensure this function complies with the
NVMe base specification, building on your patch.

Maurizio
Christoph Hellwig Nov. 22, 2024, 12:08 p.m. UTC | #3
On Fri, Nov 22, 2024 at 02:20:36PM +0530, Nilay Shroff wrote:
> The nvme_execute_identify_ns_nvm function uses ZERO_PAGE
> for copying SG list with all zeros. As ZERO_PAGE would not
> necessarily return the virtual-address of the zero page, we
> need to first convert the page address to kernel virtual-
> address and then use it as source address for copying the
> data to SG list with all zeros.
> 
> Using return address of ZERO_PAGE(0) as source address for
> copying data to SG list would fill the target buffer with
> random value and causes the undesired side effect. This patch
> implements the fix ensuring that we use virtual-address of the
> zero page for copying all zeros to the SG list buffers.

I wonder if using ZERO_PAGE() is simply a little too smart for it's
own sake and it should just use kzalloc like a bunch of other identify
implementation..
diff mbox series

Patch

diff --git a/drivers/nvme/target/admin-cmd.c b/drivers/nvme/target/admin-cmd.c
index 934b401fbc2f..a2b0444f28ab 100644
--- a/drivers/nvme/target/admin-cmd.c
+++ b/drivers/nvme/target/admin-cmd.c
@@ -901,12 +901,14 @@  static void nvmet_execute_identify_ctrl_nvm(struct nvmet_req *req)
 static void nvme_execute_identify_ns_nvm(struct nvmet_req *req)
 {
 	u16 status;
+	void *zero_buf;
 
 	status = nvmet_req_find_ns(req);
 	if (status)
 		goto out;
 
-	status = nvmet_copy_to_sgl(req, 0, ZERO_PAGE(0),
+	zero_buf = page_to_virt(ZERO_PAGE(0));
+	status = nvmet_copy_to_sgl(req, 0, zero_buf,
 				   NVME_IDENTIFY_DATA_SIZE);
 out:
 	nvmet_req_complete(req, status);