linux-kernel - Re: [PATCH RESEND] ksmbd: fix integer overflows on 32 bit systems

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <2fb3efb4-a889-4b49-8100-51147d9ae426@stanley.mountain>
Date: Tue, 14 Jan 2025 13:18:17 +0300
From: Dan Carpenter <dan.carpenter@...aro.org>
To: Namjae Jeon <linkinjeon@...nel.org>
Cc: Steve French <sfrench@...ba.org>,
	Sergey Senozhatsky <senozhatsky@...omium.org>,
	Tom Talpey <tom@...pey.com>, linux-cifs@...r.kernel.org,
	linux-kernel@...r.kernel.org, kernel-janitors@...r.kernel.org
Subject: Re: [PATCH RESEND] ksmbd: fix integer overflows on 32 bit systems

On Tue, Jan 14, 2025 at 04:53:18PM +0900, Namjae Jeon wrote:
> On Mon, Jan 13, 2025 at 3:17 PM Dan Carpenter <dan.carpenter@...aro.org> wrote:
> >
> > On 32bit systems the addition operations in ipc_msg_alloc() can
> > potentially overflow leading to memory corruption.  Fix this using
> > size_add() which will ensure that the invalid allocations do not succeed.
> You previously said that memcpy overrun does not occur due to memory
> allocation failure with SIZE_MAX.
>
> Would it be better to handle integer overflows as an error before
> memory allocation?

I mean we could do something like the below patch but I'd prefer to fix
it this way.

> And static checkers don't detect memcpy overrun by considering memory
> allocation failure?

How the struct_size()/array_size() kernel hardenning works is that if
you pass in a too large value instead of wrapping to a small value, the
math results in SIZE_MAX so the allocation will fail.  We already handle
allocation failures correctly so it's fine.

The problem in this code is that on 32 bit systems if you chose a "sz"
value which is (unsigned int)-4 then the kvzalloc() allocation will
succeed but the buffer will be 4 bytes smaller than intended and the
"msg->sz = sz;" assignment will corrupt memory.

Anyway, here is how the patch could look like with bounds checking instead
of size_add().  We could fancy it up a bit, but I don't like fancy math.

regards,
dan carpenter

diff --git a/fs/smb/server/transport_ipc.c b/fs/smb/server/transport_ipc.c
index befaf42b84cc..e1e3bfff163c 100644
--- a/fs/smb/server/transport_ipc.c
+++ b/fs/smb/server/transport_ipc.c
@@ -626,6 +626,9 @@ ksmbd_ipc_spnego_authen_request(const char *spnego_blob, int blob_len)
 	struct ksmbd_spnego_authen_request *req;
 	struct ksmbd_spnego_authen_response *resp;
 
+	if (blob_len > INT_MAX)
+		return NULL;
+
 	msg = ipc_msg_alloc(sizeof(struct ksmbd_spnego_authen_request) +
 			blob_len + 1);
 	if (!msg)
@@ -805,6 +808,9 @@ struct ksmbd_rpc_command *ksmbd_rpc_write(struct ksmbd_session *sess, int handle
 	struct ksmbd_rpc_command *req;
 	struct ksmbd_rpc_command *resp;
 
+	if (payload_sz > INT_MAX)
+		return NULL;
+
 	msg = ipc_msg_alloc(sizeof(struct ksmbd_rpc_command) + payload_sz + 1);
 	if (!msg)
 		return NULL;
@@ -853,6 +859,9 @@ struct ksmbd_rpc_command *ksmbd_rpc_ioctl(struct ksmbd_session *sess, int handle
 	struct ksmbd_rpc_command *req;
 	struct ksmbd_rpc_command *resp;
 
+	if (payload_sz > INT_MAX)
+		return NULL;
+
 	msg = ipc_msg_alloc(sizeof(struct ksmbd_rpc_command) + payload_sz + 1);
 	if (!msg)
 		return NULL;
@@ -878,6 +887,9 @@ struct ksmbd_rpc_command *ksmbd_rpc_rap(struct ksmbd_session *sess, void *payloa
 	struct ksmbd_rpc_command *req;
 	struct ksmbd_rpc_command *resp;
 
+	if (payload_sz > INT_MAX)
+		return NULL;
+
 	msg = ipc_msg_alloc(sizeof(struct ksmbd_rpc_command) + payload_sz + 1);
 	if (!msg)
 		return NULL;