[PATCH] drivers: nvme: target: core: deref after null

[PATCH] drivers: nvme: target: core: deref after null

[Alexander Sapozhnikov]

From: Alexandr Sapozhnikov <alsp705@gmail.com>

After having been compared to a NULL value at core.c:813, 
pointer '(**sq->ctrl).sqs' is dereferenced at core.c:838.

Found by Linux Verification Center (linuxtesting.org) with SVACE.

Signed-off-by: Alexandr Sapozhnikov <alsp705@gmail.com>
---
 drivers/nvme/target/core.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/nvme/target/core.c b/drivers/nvme/target/core.c
index 06f0c587f343..5a67d2bc0c55 100644
--- a/drivers/nvme/target/core.c
+++ b/drivers/nvme/target/core.c
@@ -827,7 +827,7 @@ void nvmet_sq_destroy(struct nvmet_sq *sq)
 	 */
 	ctrl = sq->ctrl;
 
-	if (ctrl) {
+	if (ctrl && ctrl->sqs) {
 		/*
 		 * The teardown flow may take some time, and the host may not
 		 * send us keep-alive during this period, hence reset the
-- 
2.39.2

Re: [PATCH] drivers: nvme: target: core: deref after null

[Chaitanya Kulkarni]

On 6/7/24 06:43, Alexander Sapozhnikov wrote:
> From: Alexandr Sapozhnikov <alsp705@gmail.com>
>
> After having been compared to a NULL value at core.c:813,
> pointer '(**sq->ctrl).sqs' is dereferenced at core.c:838.

really didn't understand the NULL check mentioned above :-

  812          */
  813         if (ctrl && ctrl->sqs && ctrl->sqs[0] == sq)
  814                 nvmet_async_events_failall(ctrl);

from what I know it checks the non NULL ctrl->sqs and then
validates the sq is admin queue before calling failing the async
events ...

> Found by Linux Verification Center (linuxtesting.org) with SVACE.
>
> Signed-off-by: Alexandr Sapozhnikov <alsp705@gmail.com>
> ---
>   drivers/nvme/target/core.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/nvme/target/core.c b/drivers/nvme/target/core.c
> index 06f0c587f343..5a67d2bc0c55 100644
> --- a/drivers/nvme/target/core.c
> +++ b/drivers/nvme/target/core.c
> @@ -827,7 +827,7 @@ void nvmet_sq_destroy(struct nvmet_sq *sq)
>   	 */
>   	ctrl = sq->ctrl;
>   
> -	if (ctrl) {
> +	if (ctrl && ctrl->sqs) {
>   		/*
>   		 * The teardown flow may take some time, and the host may not
>   		 * send us keep-alive during this period, hence reset the

do we need more fine grained check where ctrl->sqs is actually
accessed :-

diff --git a/drivers/nvme/target/core.c b/drivers/nvme/target/core.c
index 06f0c587f343..a70cad5a6262 100644
--- a/drivers/nvme/target/core.c
+++ b/drivers/nvme/target/core.c
@@ -835,7 +835,8 @@ void nvmet_sq_destroy(struct nvmet_sq *sq)
                  * controller teardown as a result of a keep-alive 
expiration.
                  */
                 ctrl->reset_tbkas = true;
-               sq->ctrl->sqs[sq->qid] = NULL;
+               if (sq->ctrl->sqs)
+                       sq->ctrl->sqs[sq->qid] = NULL;
                 nvmet_ctrl_put(ctrl);
                 sq->ctrl = NULL; /* allows reusing the queue later */
         }

-ck

Re: [PATCH] drivers: nvme: target: core: deref after null

[Markus Elfring]

> After having been compared to a NULL value at core.c:813,
> pointer '(**sq->ctrl).sqs' is dereferenced at core.c:838.

* Please choose another imperative wording for an improved change description.
  https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/process/submitting-patches.rst?h=v6.10-rc2#n94

* Would you like to add the tag “Fixes” accordingly?

* How do you think about to use the summary phrase “Avoid null pointer dereference
  in nvmet_sq_destroy()”?

* Is the subsystem specification “nvmet” more appropriate?


Regards,
Markus

Re: [PATCH] drivers: nvme: target: core: deref after null

[Sagi Grimberg]

On 07/06/2024 16:43, Alexander Sapozhnikov wrote:
> From: Alexandr Sapozhnikov <alsp705@gmail.com>
>
> After having been compared to a NULL value at core.c:813,
> pointer '(**sq->ctrl).sqs' is dereferenced at core.c:838.
>
> Found by Linux Verification Center (linuxtesting.org) with SVACE.
>
> Signed-off-by: Alexandr Sapozhnikov <alsp705@gmail.com>
> ---
>   drivers/nvme/target/core.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/nvme/target/core.c b/drivers/nvme/target/core.c
> index 06f0c587f343..5a67d2bc0c55 100644
> --- a/drivers/nvme/target/core.c
> +++ b/drivers/nvme/target/core.c
> @@ -827,7 +827,7 @@ void nvmet_sq_destroy(struct nvmet_sq *sq)
>   	 */
>   	ctrl = sq->ctrl;
>   
> -	if (ctrl) {
> +	if (ctrl && ctrl->sqs) {
>   		/*
>   		 * The teardown flow may take some time, and the host may not
>   		 * send us keep-alive during this period, hence reset the

This is a redundant check. sq->ctrl and sq->ctrl->sqs have the same 
lifetime. If
sq->ctrl was assigned, it means that it was allocated and needs a final 
put to
fully cleanup. ctrl->sqs is allocated in nvmet_alloc_ctrl and freed in 
nvmet_free_ctrl
which is the final reference.