[PATCH] PCI: prevent out of bounds access in numa_node override

[PATCH] PCI: prevent out of bounds access in numa_node override

[Sasha Levin]

Commit 63692df1 ("PCI: Allow numa_node override via sysfs") didn't check that
the numa node provided by userspace is valid. Passing a node number too high
would attempt to access invalid memory and trigger a kernel panic.

Fixes: 63692df1 ("PCI: Allow numa_node override via sysfs")
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
---
 drivers/pci/pci-sysfs.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
index 312f23a..e9abca8 100644
--- a/drivers/pci/pci-sysfs.c
+++ b/drivers/pci/pci-sysfs.c
@@ -216,7 +216,7 @@ static ssize_t numa_node_store(struct device *dev,
 	if (ret)
 		return ret;
 
-	if (!node_online(node))
+	if (node > MAX_NUMNODES || !node_online(node))
 		return -EINVAL;
 
 	add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
-- 
1.7.10.4

Re: [PATCH] PCI: prevent out of bounds access in numa_node override

[Bjorn Helgaas]

Hi Sasha,

On Sun, Oct 04, 2015 at 05:49:29PM -0400, Sasha Levin wrote:
> Commit 63692df1 ("PCI: Allow numa_node override via sysfs") didn't check that
> the numa node provided by userspace is valid. Passing a node number too high
> would attempt to access invalid memory and trigger a kernel panic.
> 
> Fixes: 63692df1 ("PCI: Allow numa_node override via sysfs")
> Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
> ---
>  drivers/pci/pci-sysfs.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
> index 312f23a..e9abca8 100644
> --- a/drivers/pci/pci-sysfs.c
> +++ b/drivers/pci/pci-sysfs.c
> @@ -216,7 +216,7 @@ static ssize_t numa_node_store(struct device *dev,
>  	if (ret)
>  		return ret;
>  
> -	if (!node_online(node))
> +	if (node > MAX_NUMNODES || !node_online(node))

This needs to be "node >= MAX_NUMNODES", doesn't it?  I'll fix it up if
you agree.

Looks like a candidate for stable.

>  		return -EINVAL;
>  
>  	add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
> -- 
> 1.7.10.4
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-pci" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] PCI: prevent out of bounds access in numa_node override

[Prarit Bhargava]

On 10/06/2015 03:36 PM, Bjorn Helgaas wrote:
> Hi Sasha,
> 
> On Sun, Oct 04, 2015 at 05:49:29PM -0400, Sasha Levin wrote:
>> Commit 63692df1 ("PCI: Allow numa_node override via sysfs") didn't check that
>> the numa node provided by userspace is valid. Passing a node number too high
>> would attempt to access invalid memory and trigger a kernel panic.
>>
>> Fixes: 63692df1 ("PCI: Allow numa_node override via sysfs")
>> Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
>> ---
>>  drivers/pci/pci-sysfs.c |    2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
>> index 312f23a..e9abca8 100644
>> --- a/drivers/pci/pci-sysfs.c
>> +++ b/drivers/pci/pci-sysfs.c
>> @@ -216,7 +216,7 @@ static ssize_t numa_node_store(struct device *dev,
>>  	if (ret)
>>  		return ret;
>>  
>> -	if (!node_online(node))
>> +	if (node > MAX_NUMNODES || !node_online(node))
> 
> This needs to be "node >= MAX_NUMNODES", doesn't it?  I'll fix it up if
> you agree.

Not a strenuous objection, but I don't see much bound checking using
MAX_NUMNODES in the kernel outside of the core numa area.  Is fixing
node_online() with bounds checking a better option here so that other callers
get the fix?  I would have thought that calling node_online() with node >
MAX_NUMNODES should be safe to call.

P.

> 
> Looks like a candidate for stable.
> 
>>  		return -EINVAL;
>>  
>>  	add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
>> -- 
>> 1.7.10.4
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-pci" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> --
> To unsubscribe from this list: send the line "unsubscribe linux-pci" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

Re: [PATCH] PCI: prevent out of bounds access in numa_node override

[Sasha Levin]

On 10/06/2015 04:02 PM, Prarit Bhargava wrote:
> 
> 
> On 10/06/2015 03:36 PM, Bjorn Helgaas wrote:
>> Hi Sasha,
>>
>> On Sun, Oct 04, 2015 at 05:49:29PM -0400, Sasha Levin wrote:
>>> Commit 63692df1 ("PCI: Allow numa_node override via sysfs") didn't check that
>>> the numa node provided by userspace is valid. Passing a node number too high
>>> would attempt to access invalid memory and trigger a kernel panic.
>>>
>>> Fixes: 63692df1 ("PCI: Allow numa_node override via sysfs")
>>> Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
>>> ---
>>>  drivers/pci/pci-sysfs.c |    2 +-
>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
>>> index 312f23a..e9abca8 100644
>>> --- a/drivers/pci/pci-sysfs.c
>>> +++ b/drivers/pci/pci-sysfs.c
>>> @@ -216,7 +216,7 @@ static ssize_t numa_node_store(struct device *dev,
>>>  	if (ret)
>>>  		return ret;
>>>  
>>> -	if (!node_online(node))
>>> +	if (node > MAX_NUMNODES || !node_online(node))
>>
>> This needs to be "node >= MAX_NUMNODES", doesn't it?  I'll fix it up if
>> you agree.

Yup, you're right.

> 
> Not a strenuous objection, but I don't see much bound checking using
> MAX_NUMNODES in the kernel outside of the core numa area.  Is fixing
> node_online() with bounds checking a better option here so that other callers
> get the fix?  I would have thought that calling node_online() with node >
> MAX_NUMNODES should be safe to call.

I don't know, this will add overhead to node_online(), and isn't really
done in any other similar function. For example, cpu_online() isn't safe
to call with cpu > NR_CPUS either.


Thanks,
Sasha

Re: [PATCH] PCI: prevent out of bounds access in numa_node override

[Bjorn Helgaas]

On Wed, Oct 07, 2015 at 10:07:33AM -0400, Sasha Levin wrote:
> > On 10/06/2015 03:36 PM, Bjorn Helgaas wrote:
> >> Hi Sasha,
> >>
> >> On Sun, Oct 04, 2015 at 05:49:29PM -0400, Sasha Levin wrote:
> >>> Commit 63692df1 ("PCI: Allow numa_node override via sysfs") didn't check that
> >>> the numa node provided by userspace is valid. Passing a node number too high
> >>> would attempt to access invalid memory and trigger a kernel panic.
> >>>
> >>> Fixes: 63692df1 ("PCI: Allow numa_node override via sysfs")
> >>> Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
> >>> ---
> >>>  drivers/pci/pci-sysfs.c |    2 +-
> >>>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>>
> >>> diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
> >>> index 312f23a..e9abca8 100644
> >>> --- a/drivers/pci/pci-sysfs.c
> >>> +++ b/drivers/pci/pci-sysfs.c
> >>> @@ -216,7 +216,7 @@ static ssize_t numa_node_store(struct device *dev,
> >>>  	if (ret)
> >>>  		return ret;
> >>>  
> >>> -	if (!node_online(node))
> >>> +	if (node > MAX_NUMNODES || !node_online(node))
> >>
> >> This needs to be "node >= MAX_NUMNODES", doesn't it?  I'll fix it up if
> >> you agree.
> 
> Yup, you're right.

OK, applied to for-linus for v4.3 as follows.  Thanks a lot, Sasha!


commit 0f7b9ad7477f7ce9b30e9ab0048f6e726f11a08a
Author: Sasha Levin <sasha.levin@oracle.com>
Date:   Sun Oct 4 17:49:29 2015 -0400

    PCI: Prevent out of bounds access in numa_node override
    
    63692df103e9 ("PCI: Allow numa_node override via sysfs") didn't check that
    the numa node provided by userspace is valid.  Passing a node number too
    high would attempt to access invalid memory and trigger a kernel panic.
    
    Fixes: 63692df103e9 ("PCI: Allow numa_node override via sysfs")
    Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
    Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
    CC: stable@vger.kernel.org	# v3.19+

diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
index 312f23a..9261868 100644
--- a/drivers/pci/pci-sysfs.c
+++ b/drivers/pci/pci-sysfs.c
@@ -216,7 +216,7 @@ static ssize_t numa_node_store(struct device *dev,
 	if (ret)
 		return ret;
 
-	if (!node_online(node))
+	if (node >= MAX_NUMNODES || !node_online(node))
 		return -EINVAL;
 
 	add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);

Re: [PATCH] PCI: prevent out of bounds access in numa_node override

[Bjorn Helgaas]

Hi Prarit,

On Tue, Oct 06, 2015 at 04:02:22PM -0400, Prarit Bhargava wrote:
> On 10/06/2015 03:36 PM, Bjorn Helgaas wrote:
> > On Sun, Oct 04, 2015 at 05:49:29PM -0400, Sasha Levin wrote:
> >> Commit 63692df1 ("PCI: Allow numa_node override via sysfs") didn't check that
> >> the numa node provided by userspace is valid. Passing a node number too high
> >> would attempt to access invalid memory and trigger a kernel panic.
> >>
> >> Fixes: 63692df1 ("PCI: Allow numa_node override via sysfs")
> >> Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
> >> ---
> >>  drivers/pci/pci-sysfs.c |    2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
> >> index 312f23a..e9abca8 100644
> >> --- a/drivers/pci/pci-sysfs.c
> >> +++ b/drivers/pci/pci-sysfs.c
> >> @@ -216,7 +216,7 @@ static ssize_t numa_node_store(struct device *dev,
> >>  	if (ret)
> >>  		return ret;
> >>  
> >> -	if (!node_online(node))
> >> +	if (node > MAX_NUMNODES || !node_online(node))
> > 
> > This needs to be "node >= MAX_NUMNODES", doesn't it?  I'll fix it up if
> > you agree.
> 
> Not a strenuous objection, but I don't see much bound checking using
> MAX_NUMNODES in the kernel outside of the core numa area.  Is fixing
> node_online() with bounds checking a better option here so that other callers
> get the fix?  I would have thought that calling node_online() with node >
> MAX_NUMNODES should be safe to call.

Yes, that would certainly be an option.  I don't feel super strongly
either way, but one argument in favor of Sasha's approach is that the
validation of user input is nice and obvious right at the point where
we process the input.

Bjorn