The validation on the length of incoming packets performed in
storvsc_on_channel_callback() does not apply to unsolicited
packets with ID of 0 sent by Hyper-V. Adjust the validation
for such unsolicited packets.
Fixes: 91b1b640b834b2 ("scsi: storvsc: Validate length of incoming packet in storvsc_on_channel_callback()")
Reported-by: Dexuan Cui <[email protected]>
Signed-off-by: Andrea Parri (Microsoft) <[email protected]>
Reviewed-by: Haiyang Zhang <[email protected]>
---
Changes since RFC[1]:
- Merge length checks (Haiyang Zhang)
[1] https://lkml.kernel.org/r/[email protected]
drivers/scsi/storvsc_drv.c | 13 ++++++++++---
1 file changed, 10 insertions(+), 3 deletions(-)
diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
index ebbbc1299c625..349c1071a98d4 100644
--- a/drivers/scsi/storvsc_drv.c
+++ b/drivers/scsi/storvsc_drv.c
@@ -292,6 +292,9 @@ struct vmstorage_protocol_version {
#define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
#define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
+/* Lower bound on the size of unsolicited packets with ID of 0 */
+#define VSTOR_MIN_UNSOL_PKT_SIZE 48
+
struct vstor_packet {
/* Requested operation type */
enum vstor_packet_operation operation;
@@ -1285,11 +1288,15 @@ static void storvsc_on_channel_callback(void *context)
foreach_vmbus_pkt(desc, channel) {
struct vstor_packet *packet = hv_pkt_data(desc);
struct storvsc_cmd_request *request = NULL;
+ u32 pktlen = hv_pkt_datalen(desc);
u64 rqst_id = desc->trans_id;
+ u32 minlen = rqst_id ? sizeof(struct vstor_packet) -
+ stor_device->vmscsi_size_delta : VSTOR_MIN_UNSOL_PKT_SIZE;
- if (hv_pkt_datalen(desc) < sizeof(struct vstor_packet) -
- stor_device->vmscsi_size_delta) {
- dev_err(&device->device, "Invalid packet len\n");
+ if (pktlen < minlen) {
+ dev_err(&device->device,
+ "Invalid pkt: id=%llu, len=%u, minlen=%u\n",
+ rqst_id, pktlen, minlen);
continue;
}
--
2.25.1
From: Andrea Parri (Microsoft) <[email protected]> Sent: Tuesday, October 5, 2021 4:41 AM
>
> The validation on the length of incoming packets performed in
> storvsc_on_channel_callback() does not apply to unsolicited
> packets with ID of 0 sent by Hyper-V. Adjust the validation
> for such unsolicited packets.
>
> Fixes: 91b1b640b834b2 ("scsi: storvsc: Validate length of incoming packet in storvsc_on_channel_callback()")
> Reported-by: Dexuan Cui <[email protected]>
> Signed-off-by: Andrea Parri (Microsoft) <[email protected]>
> Reviewed-by: Haiyang Zhang <[email protected]>
> ---
> Changes since RFC[1]:
> - Merge length checks (Haiyang Zhang)
>
> [1] https://lore.kernel.org/all/[email protected]/T/#u
>
> drivers/scsi/storvsc_drv.c | 13 ++++++++++---
> 1 file changed, 10 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
> index ebbbc1299c625..349c1071a98d4 100644
> --- a/drivers/scsi/storvsc_drv.c
> +++ b/drivers/scsi/storvsc_drv.c
> @@ -292,6 +292,9 @@ struct vmstorage_protocol_version {
> #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
> #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
>
> +/* Lower bound on the size of unsolicited packets with ID of 0 */
> +#define VSTOR_MIN_UNSOL_PKT_SIZE 48
> +
I know you have determined experimentally that Hyper-V sends
unsolicited packets with the above length, so the idea is to validate
that the guest actually gets packets at least that big. But I wonder if
we should think about this slightly differently.
The goal is for the storvsc driver to protect itself against bad or
malicious messages from Hyper-V. For the unsolicited messages, the
only field that this storvsc driver needs to access is the
vstor_packet->operation field. So an alternate approach is to set
the minimum length as small as possible while ensuring that field is valid.
Then if Hyper-V later changes the size of these unsolicited packets to
some smaller size that still contains a valid "operation" field, this code
will still work. If in a new version of the protocol Hyper-V adds fields
that this driver needs to look at, then the minimum size can be
adjusted as needed for that new protocol version.
> struct vstor_packet {
> /* Requested operation type */
> enum vstor_packet_operation operation;
> @@ -1285,11 +1288,15 @@ static void storvsc_on_channel_callback(void *context)
> foreach_vmbus_pkt(desc, channel) {
> struct vstor_packet *packet = hv_pkt_data(desc);
> struct storvsc_cmd_request *request = NULL;
> + u32 pktlen = hv_pkt_datalen(desc);
> u64 rqst_id = desc->trans_id;
> + u32 minlen = rqst_id ? sizeof(struct vstor_packet) -
> + stor_device->vmscsi_size_delta : VSTOR_MIN_UNSOL_PKT_SIZE;
>
> - if (hv_pkt_datalen(desc) < sizeof(struct vstor_packet) -
> - stor_device->vmscsi_size_delta) {
> - dev_err(&device->device, "Invalid packet len\n");
> + if (pktlen < minlen) {
> + dev_err(&device->device,
> + "Invalid pkt: id=%llu, len=%u, minlen=%u\n",
> + rqst_id, pktlen, minlen);
> continue;
> }
>
> --
> 2.25.1
I'm good with the rest of the code. It's just a question of whether to perhaps
"future-proof" the code by not requiring a packet size any bigger than the
driver actually needs to reference.
Michael
> > @@ -292,6 +292,9 @@ struct vmstorage_protocol_version {
> > #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
> > #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
> >
> > +/* Lower bound on the size of unsolicited packets with ID of 0 */
> > +#define VSTOR_MIN_UNSOL_PKT_SIZE 48
> > +
>
> I know you have determined experimentally that Hyper-V sends
> unsolicited packets with the above length, so the idea is to validate
> that the guest actually gets packets at least that big. But I wonder if
> we should think about this slightly differently.
>
> The goal is for the storvsc driver to protect itself against bad or
> malicious messages from Hyper-V. For the unsolicited messages, the
> only field that this storvsc driver needs to access is the
> vstor_packet->operation field.
Eh, this is one piece of information I was looking for... ;-)
>So an alternate approach is to set
> the minimum length as small as possible while ensuring that field is valid.
The fact is, I'm not sure how to do it for unsolicited messages.
Current code ensures/checks != COMPLETE_IO. Your comment above
and code audit suggest that we should add a check != FCHBA_DATA.
I saw ENUMERATE_BUS messages, code only using their "operation".
And, again, this is only based on current code/observations...
So, maybe you mean something like this (on top of this patch)?
diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
index 349c1071a98d4..8fedac3c7597a 100644
--- a/drivers/scsi/storvsc_drv.c
+++ b/drivers/scsi/storvsc_drv.c
@@ -292,9 +292,6 @@ struct vmstorage_protocol_version {
#define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
#define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
-/* Lower bound on the size of unsolicited packets with ID of 0 */
-#define VSTOR_MIN_UNSOL_PKT_SIZE 48
-
struct vstor_packet {
/* Requested operation type */
enum vstor_packet_operation operation;
@@ -1291,7 +1288,7 @@ static void storvsc_on_channel_callback(void *context)
u32 pktlen = hv_pkt_datalen(desc);
u64 rqst_id = desc->trans_id;
u32 minlen = rqst_id ? sizeof(struct vstor_packet) -
- stor_device->vmscsi_size_delta : VSTOR_MIN_UNSOL_PKT_SIZE;
+ stor_device->vmscsi_size_delta : sizeof(enum vstor_packet_operation);
if (pktlen < minlen) {
dev_err(&device->device,
@@ -1315,7 +1312,8 @@ static void storvsc_on_channel_callback(void *context)
* storvsc_on_io_completion() with a guest memory address that is
* zero if Hyper-V were to construct and send such a bogus packet.
*/
- if (packet->operation == VSTOR_OPERATION_COMPLETE_IO) {
+ if (packet->operation == VSTOR_OPERATION_COMPLETE_IO ||
+ packet->operation == VSTOR_OPERATION_FCHBA_DATA) {
dev_err(&device->device, "Invalid packet with ID of 0\n");
continue;
}
Thanks,
Andrea
From: Andrea Parri <[email protected]> Sent: Tuesday, October 5, 2021 11:14 AM
>
> > > @@ -292,6 +292,9 @@ struct vmstorage_protocol_version {
> > > #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
> > > #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
> > >
> > > +/* Lower bound on the size of unsolicited packets with ID of 0 */
> > > +#define VSTOR_MIN_UNSOL_PKT_SIZE 48
> > > +
> >
> > I know you have determined experimentally that Hyper-V sends
> > unsolicited packets with the above length, so the idea is to validate
> > that the guest actually gets packets at least that big. But I wonder if
> > we should think about this slightly differently.
> >
> > The goal is for the storvsc driver to protect itself against bad or
> > malicious messages from Hyper-V. For the unsolicited messages, the
> > only field that this storvsc driver needs to access is the
> > vstor_packet->operation field.
>
> Eh, this is one piece of information I was looking for... ;-)
I'm just looking at the code in storvsc_on_receive(). storvsc_on_receive()
itself looks at the "operation" field, but for the REMOVE_DEVICE and
ENUMERATE_BUS operations, you can see that the rest of the vstor_packet
is ignored and is not passed to any called functions.
>
>
> >So an alternate approach is to set
> > the minimum length as small as possible while ensuring that field is valid.
>
> The fact is, I'm not sure how to do it for unsolicited messages.
> Current code ensures/checks != COMPLETE_IO. Your comment above
> and code audit suggest that we should add a check != FCHBA_DATA.
> I saw ENUMERATE_BUS messages, code only using their "operation".
I'm not completely sure about FCHBA_DATA. That message does not
seem to be unsolicited, as the guest sends out a message of that type in
storvsc_channel_init() using storvsc_execute_vstor_op(). So any received
messages of that type are presumably in response to the guest request,
and will get handled via the test for rqst_id == VMBUS_RQST_INIT. Long
Li could probably confirm. So if Hyper-V did send a FCHBA_DATA
packet with rqst_id of 0, it would seem to be appropriate to reject
it.
>
> And, again, this is only based on current code/observations...
>
> So, maybe you mean something like this (on top of this patch)?
Yes, with a comment to explain what's going on. :-)
>
> diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
> index 349c1071a98d4..8fedac3c7597a 100644
> --- a/drivers/scsi/storvsc_drv.c
> +++ b/drivers/scsi/storvsc_drv.c
> @@ -292,9 +292,6 @@ struct vmstorage_protocol_version {
> #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
> #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
>
> -/* Lower bound on the size of unsolicited packets with ID of 0 */
> -#define VSTOR_MIN_UNSOL_PKT_SIZE 48
> -
> struct vstor_packet {
> /* Requested operation type */
> enum vstor_packet_operation operation;
> @@ -1291,7 +1288,7 @@ static void storvsc_on_channel_callback(void *context)
> u32 pktlen = hv_pkt_datalen(desc);
> u64 rqst_id = desc->trans_id;
> u32 minlen = rqst_id ? sizeof(struct vstor_packet) -
> - stor_device->vmscsi_size_delta : VSTOR_MIN_UNSOL_PKT_SIZE;
> + stor_device->vmscsi_size_delta : sizeof(enum vstor_packet_operation);
>
> if (pktlen < minlen) {
> dev_err(&device->device,
> @@ -1315,7 +1312,8 @@ static void storvsc_on_channel_callback(void *context)
> * storvsc_on_io_completion() with a guest memory address that is
> * zero if Hyper-V were to construct and send such a bogus packet.
> */
> - if (packet->operation == VSTOR_OPERATION_COMPLETE_IO) {
> + if (packet->operation == VSTOR_OPERATION_COMPLETE_IO ||
> + packet->operation == VSTOR_OPERATION_FCHBA_DATA) {
> dev_err(&device->device, "Invalid packet with ID of 0\n");
> continue;
> }
>
> Thanks,
> Andrea
> > > I know you have determined experimentally that Hyper-V sends
> > > unsolicited packets with the above length, so the idea is to validate
> > > that the guest actually gets packets at least that big. But I wonder if
> > > we should think about this slightly differently.
> > >
> > > The goal is for the storvsc driver to protect itself against bad or
> > > malicious messages from Hyper-V. For the unsolicited messages, the
> > > only field that this storvsc driver needs to access is the
> > > vstor_packet->operation field.
> >
> > Eh, this is one piece of information I was looking for... ;-)
>
> I'm just looking at the code in storvsc_on_receive(). storvsc_on_receive()
> itself looks at the "operation" field, but for the REMOVE_DEVICE and
> ENUMERATE_BUS operations, you can see that the rest of the vstor_packet
> is ignored and is not passed to any called functions.
>
> >
> >
> > >So an alternate approach is to set
> > > the minimum length as small as possible while ensuring that field is valid.
> >
> > The fact is, I'm not sure how to do it for unsolicited messages.
> > Current code ensures/checks != COMPLETE_IO. Your comment above
> > and code audit suggest that we should add a check != FCHBA_DATA.
> > I saw ENUMERATE_BUS messages, code only using their "operation".
>
> I'm not completely sure about FCHBA_DATA. That message does not
> seem to be unsolicited, as the guest sends out a message of that type in
> storvsc_channel_init() using storvsc_execute_vstor_op(). So any received
> messages of that type are presumably in response to the guest request,
> and will get handled via the test for rqst_id == VMBUS_RQST_INIT. Long
> Li could probably confirm. So if Hyper-V did send a FCHBA_DATA
> packet with rqst_id of 0, it would seem to be appropriate to reject
> it.
>
> >
> > And, again, this is only based on current code/observations...
> >
> > So, maybe you mean something like this (on top of this patch)?
>
> Yes, with a comment to explain what's going on. :-)
My (current) best guess is here:
https://lkml.kernel.org/r/[email protected]
Thanks,
Andrea