[PATCH] Drivers: hv: vmbus: Initialize unload_event statically

[PATCH] Drivers: hv: vmbus: Initialize unload_event statically

[Andrea Parri (Microsoft)]

If a malicious or compromised Hyper-V sends a spurious message of type
CHANNELMSG_UNLOAD_RESPONSE, the function vmbus_unload_response() will
call complete() on an uninitialized event, and cause an oops.

Reported-by: Michael Kelley <mikelley@microsoft.com>
Signed-off-by: Andrea Parri (Microsoft) <parri.andrea@gmail.com>
---
 drivers/hv/channel_mgmt.c | 2 +-
 drivers/hv/connection.c   | 2 ++
 2 files changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/hv/channel_mgmt.c b/drivers/hv/channel_mgmt.c
index f3cf4af01e102..1efb616480a64 100644
--- a/drivers/hv/channel_mgmt.c
+++ b/drivers/hv/channel_mgmt.c
@@ -841,7 +841,7 @@ void vmbus_initiate_unload(bool crash)
 	if (vmbus_proto_version < VERSION_WIN8_1)
 		return;
 
-	init_completion(&vmbus_connection.unload_event);
+	reinit_completion(&vmbus_connection.unload_event);
 	memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
 	hdr.msgtype = CHANNELMSG_UNLOAD;
 	vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
diff --git a/drivers/hv/connection.c b/drivers/hv/connection.c
index 350e8c5cafa8c..529dcc47f3e11 100644
--- a/drivers/hv/connection.c
+++ b/drivers/hv/connection.c
@@ -26,6 +26,8 @@
 
 struct vmbus_connection vmbus_connection = {
 	.conn_state		= DISCONNECTED,
+	.unload_event		= COMPLETION_INITIALIZER(
+				  vmbus_connection.unload_event),
 	.next_gpadl_handle	= ATOMIC_INIT(0xE1E10),
 
 	.ready_for_suspend_event = COMPLETION_INITIALIZER(
-- 
2.25.1

RE: [PATCH] Drivers: hv: vmbus: Initialize unload_event statically

[Michael Kelley]

From: Andrea Parri (Microsoft) <parri.andrea@gmail.com> Sent: Friday, April 16, 2021 7:40 AM
> 
> If a malicious or compromised Hyper-V sends a spurious message of type
> CHANNELMSG_UNLOAD_RESPONSE, the function vmbus_unload_response() will
> call complete() on an uninitialized event, and cause an oops.

Please leave a comment somewhere in the code itself that describes this
scenario so that somebody in the future doesn't decide it's OK to "simplify" the
initialization of unload_event. :-)

Michael

> 
> Reported-by: Michael Kelley <mikelley@microsoft.com>
> Signed-off-by: Andrea Parri (Microsoft) <parri.andrea@gmail.com>
> ---
>  drivers/hv/channel_mgmt.c | 2 +-
>  drivers/hv/connection.c   | 2 ++
>  2 files changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/hv/channel_mgmt.c b/drivers/hv/channel_mgmt.c
> index f3cf4af01e102..1efb616480a64 100644
> --- a/drivers/hv/channel_mgmt.c
> +++ b/drivers/hv/channel_mgmt.c
> @@ -841,7 +841,7 @@ void vmbus_initiate_unload(bool crash)
>  	if (vmbus_proto_version < VERSION_WIN8_1)
>  		return;
> 
> -	init_completion(&vmbus_connection.unload_event);
> +	reinit_completion(&vmbus_connection.unload_event);
>  	memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
>  	hdr.msgtype = CHANNELMSG_UNLOAD;
>  	vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
> diff --git a/drivers/hv/connection.c b/drivers/hv/connection.c
> index 350e8c5cafa8c..529dcc47f3e11 100644
> --- a/drivers/hv/connection.c
> +++ b/drivers/hv/connection.c
> @@ -26,6 +26,8 @@
> 
>  struct vmbus_connection vmbus_connection = {
>  	.conn_state		= DISCONNECTED,
> +	.unload_event		= COMPLETION_INITIALIZER(
> +				  vmbus_connection.unload_event),
>  	.next_gpadl_handle	= ATOMIC_INIT(0xE1E10),
> 
>  	.ready_for_suspend_event = COMPLETION_INITIALIZER(
> --
> 2.25.1

Re: [PATCH] Drivers: hv: vmbus: Initialize unload_event statically

[Andrea Parri]

On Fri, Apr 16, 2021 at 03:25:03PM +0000, Michael Kelley wrote:
> From: Andrea Parri (Microsoft) <parri.andrea@gmail.com> Sent: Friday, April 16, 2021 7:40 AM
> > 
> > If a malicious or compromised Hyper-V sends a spurious message of type
> > CHANNELMSG_UNLOAD_RESPONSE, the function vmbus_unload_response() will
> > call complete() on an uninitialized event, and cause an oops.
> 
> Please leave a comment somewhere in the code itself that describes this
> scenario so that somebody in the future doesn't decide it's OK to "simplify" the
> initialization of unload_event. :-)

Yes, will do for v2.

Thanks,
  Andrea