[v1] ksmbd: fix leak of transform buffer on encrypt_resp() failure

[PATCH] ksmbd: fix leak of transform buffer on encrypt_resp() failure

Posted by Qianchang Zhao 1 month, 2 weeks ago

When encrypt_resp() fails at the send path, we only set
STATUS_DATA_ERROR but leave the transform buffer allocated (work->tr_buf
in this tree). Repeating this path leaks kernel memory and can lead to
OOM (DoS) when encryption is required.

Reproduced on: Linux v6.18-rc2 (self-built test kernel)

Fix by freeing the transform buffer and forcing plaintext error reply.

Reported-by: Qianchang Zhao <pioooooooooip@gmail.com>
Reported-by: Zhitong Liu <liuzhitong1993@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Qianchang Zhao <pioooooooooip@gmail.com>
---
 fs/smb/server/server.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/fs/smb/server/server.c b/fs/smb/server/server.c
index 40420544c..15dd13e76 100644
--- a/fs/smb/server/server.c
+++ b/fs/smb/server/server.c
@@ -244,8 +244,14 @@ static void __handle_ksmbd_work(struct ksmbd_work *work,
 	if (work->sess && work->sess->enc && work->encrypted &&
 	    conn->ops->encrypt_resp) {
 		rc = conn->ops->encrypt_resp(work);
-		if (rc < 0)
+		if (rc < 0) {
 			conn->ops->set_rsp_status(work, STATUS_DATA_ERROR);
+			work->encrypted = false;
+			if (work->tr_buf) {
+				kvfree(work->tr_buf);
+				work->tr_buf = NULL;
+			}
+		}
 	}
 	if (work->sess)
 		ksmbd_user_session_put(work->sess);
-- 
2.34.1

Re: [PATCH] ksmbd: fix leak of transform buffer on encrypt_resp() failure

Posted by Namjae Jeon 1 month, 2 weeks ago

On Tue, Nov 4, 2025 at 11:12 PM Qianchang Zhao <pioooooooooip@gmail.com> wrote:
>
> When encrypt_resp() fails at the send path, we only set
> STATUS_DATA_ERROR but leave the transform buffer allocated (work->tr_buf
> in this tree). Repeating this path leaks kernel memory and can lead to
> OOM (DoS) when encryption is required.
>
> Reproduced on: Linux v6.18-rc2 (self-built test kernel)
>
> Fix by freeing the transform buffer and forcing plaintext error reply.
>
> Reported-by: Qianchang Zhao <pioooooooooip@gmail.com>
> Reported-by: Zhitong Liu <liuzhitong1993@gmail.com>
> Cc: stable@vger.kernel.org
> Signed-off-by: Qianchang Zhao <pioooooooooip@gmail.com>
> ---
>  fs/smb/server/server.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/fs/smb/server/server.c b/fs/smb/server/server.c
> index 40420544c..15dd13e76 100644
> --- a/fs/smb/server/server.c
> +++ b/fs/smb/server/server.c
> @@ -244,8 +244,14 @@ static void __handle_ksmbd_work(struct ksmbd_work *work,
>         if (work->sess && work->sess->enc && work->encrypted &&
>             conn->ops->encrypt_resp) {
>                 rc = conn->ops->encrypt_resp(work);
> -               if (rc < 0)
> +               if (rc < 0) {
>                         conn->ops->set_rsp_status(work, STATUS_DATA_ERROR);
> +                       work->encrypted = false;
> +                       if (work->tr_buf) {
> +                               kvfree(work->tr_buf);
->tr_buf is freed in ksmbd_free_work_struct(). How can tr_buf not be freed?
Thanks.
> +                               work->tr_buf = NULL;
> +                       }
> +               }
>         }
>         if (work->sess)
>                 ksmbd_user_session_put(work->sess);
> --
> 2.34.1
>

[PATCH v2] ksmbd: clear 'encrypted' on encrypt_resp() failure to send plaintext error

Posted by Qianchang Zhao 1 month, 1 week ago

When encrypt_resp() fails in the send path, we set STATUS_DATA_ERROR but
leave work->encrypted true. The send path then still assumes a valid
transform buffer and tries to build/send an encrypted reply.

Clear work->encrypted on failure to force a plaintext error reply.
The transform buffer (if allocated) is released by ksmbd_free_work_struct(),
so no explicit kvfree(tr_buf) is needed.

Reported-by: Qianchang Zhao <pioooooooooip@gmail.com>
Reported-by: Zhitong Liu <liuzhitong1993@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Qianchang Zhao <pioooooooooip@gmail.com>

---
v2:
  - Drop explicit kvfree(tr_buf); it is freed in ksmbd_free_work_struct().
  - Keep only 'work->encrypted = false' and update the commit message.

 fs/smb/server/server.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/fs/smb/server/server.c b/fs/smb/server/server.c
index 40420544c..a7444a78f 100644
--- a/fs/smb/server/server.c
+++ b/fs/smb/server/server.c
@@ -244,8 +244,10 @@ static void __handle_ksmbd_work(struct ksmbd_work *work,
 	if (work->sess && work->sess->enc && work->encrypted &&
 	    conn->ops->encrypt_resp) {
 		rc = conn->ops->encrypt_resp(work);
-		if (rc < 0)
+		if (rc < 0) {
 			conn->ops->set_rsp_status(work, STATUS_DATA_ERROR);
+			work->encrypted = false;
+		}
 	}
 	if (work->sess)
 		ksmbd_user_session_put(work->sess);
-- 
2.34.1

Re: [PATCH v2] ksmbd: clear 'encrypted' on encrypt_resp() failure to send plaintext error

Posted by Namjae Jeon 1 month, 1 week ago

On Thu, Nov 6, 2025 at 3:58 PM Qianchang Zhao <pioooooooooip@gmail.com> wrote:
>
> When encrypt_resp() fails in the send path, we set STATUS_DATA_ERROR but
> leave work->encrypted true. The send path then still assumes a valid
> transform buffer and tries to build/send an encrypted reply.
>
> Clear work->encrypted on failure to force a plaintext error reply.
ksmbd will send a plain text error reply regardless of the
work->encrypted value.
It doesn't seem to make any sense to set this to false.
Thanks.