[PATCH net] mlx4: Fix information leak on failure to read module EEPROM

[PATCH net] mlx4: Fix information leak on failure to read module EEPROM

[Ben Hutchings]

[-- Attachment #1: Type: text/plain, Size: 1680 bytes --]

mlx4_en_get_module_eeprom() returns 0 even if it fails.  This results
in copying an uninitialised (or partly initialised) buffer back to
user-space.

Change it so that:

* In the special case that the DOM turns out not to be readable, the
  remaining part of the buffer is cleared.  This should avoid a
  regression when reading modules with this problem.

* In other error cases, the error code is propagated.

Reported-by: Yannis Aribaud <bugs@d6bell.net>
References: https://bugs.debian.org/960702
Fixes: 7202da8b7f71 ("ethtool, net/mlx4_en: Cable info, get_module_info/...")
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
---
This is compile-tested only.  It should go to stable, if it is a
correct fix.

Ben.

 drivers/net/ethernet/mellanox/mlx4/en_ethtool.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c b/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
index 8a5ea2543670..6edc3177af1c 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
@@ -2078,14 +2078,17 @@ static int mlx4_en_get_module_eeprom(struct net_device *dev,
 		ret = mlx4_get_module_info(mdev->dev, priv->port,
 					   offset, ee->len - i, data + i);
 
-		if (!ret) /* Done reading */
+		if (!ret) {
+			/* DOM was not readable after all */
+			memset(data + i, 0, ee->len - i);
 			return 0;
+		}
 
 		if (ret < 0) {
 			en_err(priv,
 			       "mlx4_get_module_info i(%d) offset(%d) bytes_to_read(%d) - FAILED (0x%x)\n",
 			       i, offset, ee->len - i, ret);
-			return 0;
+			return ret;
 		}
 
 		i += ret;

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH net] mlx4: Fix information leak on failure to read module EEPROM

[Saeed Mahameed]

On Sun, 2020-05-17 at 18:20 +0100, Ben Hutchings wrote:
> mlx4_en_get_module_eeprom() returns 0 even if it fails.  This results
> in copying an uninitialised (or partly initialised) buffer back to
> user-space.
> 
> Change it so that:
> 
> * In the special case that the DOM turns out not to be readable, the
>   remaining part of the buffer is cleared.  This should avoid a
>   regression when reading modules with this problem.
> 
> * In other error cases, the error code is propagated.
> 
> Reported-by: Yannis Aribaud <bugs@d6bell.net>
> References: https://bugs.debian.org/960702
> Fixes: 7202da8b7f71 ("ethtool, net/mlx4_en: Cable info,
> get_module_info/...")
> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
> ---
> This is compile-tested only.  It should go to stable, if it is a
> correct fix.
> 
> Ben.
> 
>  drivers/net/ethernet/mellanox/mlx4/en_ethtool.c | 7 +++++--
>  1 file changed, 5 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
> b/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
> index 8a5ea2543670..6edc3177af1c 100644
> --- a/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
> +++ b/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
> @@ -2078,14 +2078,17 @@ static int mlx4_en_get_module_eeprom(struct
> net_device *dev,
>  		ret = mlx4_get_module_info(mdev->dev, priv->port,
>  					   offset, ee->len - i, data +
> i);
>  

I am not sure i see the issue in here, and why we need the partial
memset ?
first thing in this function we do:
memset(data, 0, ee->len);

and then mlx4_get_module_info() will only copy valid data only on
success.


> -		if (!ret) /* Done reading */
> +		if (!ret) {
> +			/* DOM was not readable after all */

actually if mlx4_get_module_info()  returns any non-negative value it
means how much data was read, so if it returns 0, it means that this
was the last iteration and we are done reading the eeprom.. 

so i would remove the above comment and the memset below is redundant
since we already memset the whole buffer before the while loop.

> +			memset(data + i, 0, ee->len - i);
>  			return 0;
> +		}
>  
>  		if (ret < 0) {
>  			en_err(priv,
>  			       "mlx4_get_module_info i(%d) offset(%d)
> bytes_to_read(%d) - FAILED (0x%x)\n",
>  			       i, offset, ee->len - i, ret);
> -			return 0;
> +			return ret;

I think returning error in here was the actual solution for your
problem. you can verify by looking in the kernel log and verify you see
the log message.

>  		}
>  
>  		i += ret;

Re: [PATCH net] mlx4: Fix information leak on failure to read module EEPROM

[Ben Hutchings]

[-- Attachment #1: Type: text/plain, Size: 2355 bytes --]

On Mon, 2020-05-18 at 16:47 +0000, Saeed Mahameed wrote:
> On Sun, 2020-05-17 at 18:20 +0100, Ben Hutchings wrote:
> > mlx4_en_get_module_eeprom() returns 0 even if it fails.  This results
> > in copying an uninitialised (or partly initialised) buffer back to
> > user-space.
[...]
> I am not sure i see the issue in here, and why we need the partial
> memset ?
> first thing in this function we do:
> memset(data, 0, ee->len);
> 
> and then mlx4_get_module_info() will only copy valid data only on
> success.

Wow, sorry, I don't know how I missed that.  So this is not the bug I
was looking for.

> 
> > -		if (!ret) /* Done reading */
> > +		if (!ret) {
> > +			/* DOM was not readable after all */
> 
> actually if mlx4_get_module_info()  returns any non-negative value it
> means how much data was read, so if it returns 0, it means that this
> was the last iteration and we are done reading the eeprom.. 
> 
> so i would remove the above comment and the memset below is redundant
> since we already memset the whole buffer before the while loop.

Right.

> > +			memset(data + i, 0, ee->len - i);
> >  			return 0;
> > +		}
> >  
> >  		if (ret < 0) {
> >  			en_err(priv,
> >  			       "mlx4_get_module_info i(%d) offset(%d)
> > bytes_to_read(%d) - FAILED (0x%x)\n",
> >  			       i, offset, ee->len - i, ret);
> > -			return 0;
> > +			return ret;
> 
> I think returning error in here was the actual solution for your
> problem. you can verify by looking in the kernel log and verify you see
> the log message.

The original bug report (https://bugs.debian.org/960702) says that
ethtool reports different values depending on whether its output is
redirected.  Although returning all-zeroes for the unreadable part
might be wrong, it doesn't explain that behaviour.

Perhaps if the timing of the I²C reads is marginal, varying numbers of
bytes of DOM information might be readable?  But I don't see how
redirection of ethtool's output would affect that.  It uses a single
ioctl to read everything, and the kernel controls timing within that.

So I am mystified about what is going on here.  Maybe there is a bug in
ethtool, but I'm not seeing it.

Ben.

> >  		}
> >  
> >  		i += ret;
-- 
Ben Hutchings
The two most common things in the universe are hydrogen and stupidity.


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 833 bytes --]