PCI/switchtec: Fix Spectre v1 vulnerability

The "partition" member of the struct switchtec_ioctl_pff_port can be
indirectly controlled from user-space through an IOCTL that the device
driver provides enabling conversion between a PCI Function Framework
(PFF) number and Switchtec logical port ID and partition number, thus
allowing for command-line tooling [1] interact with the device from
user-space.

This can lead to potential exploitation of the Spectre variant 1 [2]
vulnerability since the value of the partition is then used directly
as an index to mmio_part_cfg_all of the struct switchtec_dev to retrieve
configuration from Switchtec for a specific partition number.

Fix this by sanitizing the value coming from user-space through the
available IOCTL before it's then used as an index to mmio_part_cfg_all.

This issue was detected with the help of Smatch:

  drivers/pci/switch/switchtec.c:1118 ioctl_port_to_pff() warn:
  potential spectre issue 'stdev->mmio_part_cfg_all' [r] (local cap)

Notice that given that speculation windows are large, the policy is
to kill the speculation on the first load and not worry if it can be
completed with a dependent load/store [3].

Related commit 46feb6b495f7 ("switchtec: Fix Spectre v1 vulnerability").

1. https://github.com/Microsemi/switchtec-user/blob/master/lib/platform/linux.c
2. https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/spectre.html
3. https://lore.kernel.org/lkml/CAPcyv4gLKYiCtXsKFX2FY+rW93aRtQt9zB8hU1hMsj770m8gxQ@mail.gmail.com/

Signed-off-by: Krzysztof Wilczyński <kw@linux.com>
---
 drivers/pci/switch/switchtec.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

On 2021-02-19 11:28 p.m., Krzysztof Wilczyński wrote:
> The "partition" member of the struct switchtec_ioctl_pff_port can be
> indirectly controlled from user-space through an IOCTL that the device
> driver provides enabling conversion between a PCI Function Framework
> (PFF) number and Switchtec logical port ID and partition number, thus
> allowing for command-line tooling [1] interact with the device from
> user-space.
> 
> This can lead to potential exploitation of the Spectre variant 1 [2]
> vulnerability since the value of the partition is then used directly
> as an index to mmio_part_cfg_all of the struct switchtec_dev to retrieve
> configuration from Switchtec for a specific partition number.
> 
> Fix this by sanitizing the value coming from user-space through the
> available IOCTL before it's then used as an index to mmio_part_cfg_all.
> 
> This issue was detected with the help of Smatch:
> 
>   drivers/pci/switch/switchtec.c:1118 ioctl_port_to_pff() warn:
>   potential spectre issue 'stdev->mmio_part_cfg_all' [r] (local cap)
> 
> Notice that given that speculation windows are large, the policy is
> to kill the speculation on the first load and not worry if it can be
> completed with a dependent load/store [3].
> 
> Related commit 46feb6b495f7 ("switchtec: Fix Spectre v1 vulnerability").
> 
> 1. https://github.com/Microsemi/switchtec-user/blob/master/lib/platform/linux.c
> 2. https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/spectre.html
> 3. https://lore.kernel.org/lkml/CAPcyv4gLKYiCtXsKFX2FY+rW93aRtQt9zB8hU1hMsj770m8gxQ@mail.gmail.com/
> 
> Signed-off-by: Krzysztof Wilczyński <kw@linux.com>

Looks good to me.

Reviewed-by: Logan Gunthorpe <logang@deltatee.com>

Thanks!

Logan

On Sat, Feb 20, 2021 at 06:28:37AM +0000, Krzysztof Wilczyński wrote:
> The "partition" member of the struct switchtec_ioctl_pff_port can be
> indirectly controlled from user-space through an IOCTL that the device
> driver provides enabling conversion between a PCI Function Framework
> (PFF) number and Switchtec logical port ID and partition number, thus
> allowing for command-line tooling [1] interact with the device from
> user-space.
> 
> This can lead to potential exploitation of the Spectre variant 1 [2]
> vulnerability since the value of the partition is then used directly
> as an index to mmio_part_cfg_all of the struct switchtec_dev to retrieve
> configuration from Switchtec for a specific partition number.
> 
> Fix this by sanitizing the value coming from user-space through the
> available IOCTL before it's then used as an index to mmio_part_cfg_all.
> 
> This issue was detected with the help of Smatch:
> 
>   drivers/pci/switch/switchtec.c:1118 ioctl_port_to_pff() warn:
>   potential spectre issue 'stdev->mmio_part_cfg_all' [r] (local cap)
> 
> Notice that given that speculation windows are large, the policy is
> to kill the speculation on the first load and not worry if it can be
> completed with a dependent load/store [3].
> 
> Related commit 46feb6b495f7 ("switchtec: Fix Spectre v1 vulnerability").
> 
> 1. https://github.com/Microsemi/switchtec-user/blob/master/lib/platform/linux.c
> 2. https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/spectre.html
> 3. https://lore.kernel.org/lkml/CAPcyv4gLKYiCtXsKFX2FY+rW93aRtQt9zB8hU1hMsj770m8gxQ@mail.gmail.com/
> 
> Signed-off-by: Krzysztof Wilczyński <kw@linux.com>

Applied with Logan's ack to for-linus for v5.12, thanks!

> ---
>  drivers/pci/switch/switchtec.c | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/pci/switch/switchtec.c b/drivers/pci/switch/switchtec.c
> index ba52459928f7..bb6957101fc0 100644
> --- a/drivers/pci/switch/switchtec.c
> +++ b/drivers/pci/switch/switchtec.c
> @@ -1112,12 +1112,15 @@ static int ioctl_port_to_pff(struct switchtec_dev *stdev,
>  	if (copy_from_user(&p, up, sizeof(p)))
>  		return -EFAULT;
>  
> -	if (p.partition == SWITCHTEC_IOCTL_EVENT_LOCAL_PART_IDX)
> +	if (p.partition == SWITCHTEC_IOCTL_EVENT_LOCAL_PART_IDX) {
>  		pcfg = stdev->mmio_part_cfg;
> -	else if (p.partition < stdev->partition_count)
> +	} else if (p.partition < stdev->partition_count) {
> +		p.partition = array_index_nospec(p.partition,
> +						 stdev->partition_count);
>  		pcfg = &stdev->mmio_part_cfg_all[p.partition];
> -	else
> +	} else {
>  		return -EINVAL;
> +	}
>  
>  	switch (p.port) {
>  	case 0:
> -- 
> 2.30.0
>