RE: [PATCH] VMCI: Verify PPNs before sending to device

[Jorgen S. Hansen]

> -----Original Message-----
> From: Jorgen Hansen [mailto:jhansen@xxxxxxxxxx]
> Sent: Friday, January 11, 2019 8:58 AM
> To: linux-kernel@xxxxxxxxxxxxxxx; virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
> Cc: gregkh@xxxxxxxxxxxxxxxxxxx; Pv-drivers <Pv-drivers@xxxxxxxxxx>;
> Jorgen S. Hansen <jhansen@xxxxxxxxxx>
> Subject: [PATCH] VMCI: Verify PPNs before sending to device
> 
> The current version of the VMCI device only supports 32 bit PPNs, so check
> whether we are truncating PPNs, and fail the operation if we do. One such
> check did exist, but was wrong. Another check was missing.
> 
> Testing through code modification: constructed PPN not representable by
> 32-bit and observed that failure was reported.
> 
> Fixes: 1f166439917b ("VMCI: guest side driver implementation.")
> Fixes: 06164d2b72aa ("VMCI: queue pairs implementation.")
> 
> Signed-off-by: Jorgen Hansen <jhansen@xxxxxxxxxx>
> Reviewed-by: Adit Ranadive <aditr@xxxxxxxxxx>
> Reviewed-by: Vishnu Dasa <vdasa@xxxxxxxxxx>
> ---
>  drivers/misc/vmw_vmci/vmci_guest.c      | 10 +++++++---
>  drivers/misc/vmw_vmci/vmci_queue_pair.c | 10 ++++------
>  2 files changed, 11 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/misc/vmw_vmci/vmci_guest.c
> b/drivers/misc/vmw_vmci/vmci_guest.c
> index dad5abee656e..02bb3866cf9e 100644
> --- a/drivers/misc/vmw_vmci/vmci_guest.c
> +++ b/drivers/misc/vmw_vmci/vmci_guest.c
> @@ -532,10 +532,14 @@ static int vmci_guest_probe_device(struct pci_dev
> *pdev,
>  	if (capabilities & VMCI_CAPS_NOTIFICATIONS) {
>  		unsigned long bitmap_ppn =
>  			vmci_dev->notification_base >> PAGE_SHIFT;
> -		if (!vmci_dbell_register_notification_bitmap(bitmap_ppn)) {
> +		u32 bitmap_ppn32 = bitmap_ppn;
> +
> +		if ((sizeof(bitmap_ppn) > sizeof(bitmap_ppn32)
> +		     && bitmap_ppn != bitmap_ppn32) ||
> +		    !vmci_dbell_register_notification_bitmap(bitmap_ppn)) {
>  			dev_warn(&pdev->dev,
> -				 "VMCI device unable to register notification
> bitmap with PPN 0x%x\n",
> -				 (u32) bitmap_ppn);
> +				 "VMCI device unable to register notification
> bitmap with PPN 0x%lx\n",
> +				 bitmap_ppn);
>  			error = -ENXIO;
>  			goto err_remove_vmci_dev_g;
>  		}
> diff --git a/drivers/misc/vmw_vmci/vmci_queue_pair.c
> b/drivers/misc/vmw_vmci/vmci_queue_pair.c
> index 264f4ed8eef2..1da4f6cb01b2 100644
> --- a/drivers/misc/vmw_vmci/vmci_queue_pair.c
> +++ b/drivers/misc/vmw_vmci/vmci_queue_pair.c
> @@ -465,9 +465,8 @@ static int qp_alloc_ppn_set(void *prod_q,
>  	for (i = 0; i < num_produce_pages; i++) {
>  		unsigned long pfn;
> 
> -		produce_ppns[i] =
> -			produce_q->kernel_if->u.g.pas[i] >> PAGE_SHIFT;
> -		pfn = produce_ppns[i];
> +		pfn = produce_q->kernel_if->u.g.pas[i] >> PAGE_SHIFT;
> +		produce_ppns[i] = pfn;
> 
>  		/* Fail allocation if PFN isn't supported by hypervisor. */
>  		if (sizeof(pfn) > sizeof(*produce_ppns) @@ -478,9 +477,8
> @@ static int qp_alloc_ppn_set(void *prod_q,
>  	for (i = 0; i < num_consume_pages; i++) {
>  		unsigned long pfn;
> 
> -		consume_ppns[i] =
> -			consume_q->kernel_if->u.g.pas[i] >> PAGE_SHIFT;
> -		pfn = consume_ppns[i];
> +		pfn = consume_q->kernel_if->u.g.pas[i] >> PAGE_SHIFT;
> +		consume_ppns[i] = pfn;
> 
>  		/* Fail allocation if PFN isn't supported by hypervisor. */
>  		if (sizeof(pfn) > sizeof(*consume_ppns)
> --
> 2.17.1

Please ignore this patch.

Thomas Hellstrom reminded me that the dma mask for the device should
be able to take care of this and since the VMCI device uses the default
DMA mask, the device should only get 32-bit addresses. So we should
Just remove the (wrong) overflow checks in vmci_queue_pair.c

Another patch supporting larger addresses will be sent out later.

Thanks,
Jorgen