[PATCH] Fix memory leak in mtd_dataflash

[H Hartley Sweeten]

On Sunday, October 11, 2009 6:10 AM, Artem Bityutskiy wrote:
> On Wed, 2009-10-07 at 17:08 -0400, H Hartley Sweeten wrote:
>> Fix a potential memory leak in mtd_dataflash driver.
>> 
>> The private data that is allocated when registering a DataFlash
>> device with the MTD subsystem is not released if an error occurs
>> when add_mtd_partitions() or add_mtd_device() is called.  Fix this
>> by adding an error path.  The memory is already released during a
>> remove.
>> 
>> Also, add a dev_set_drvdata(&spi->dev, NULL) before the kfree() so
>> that the spi device does not reference invalid data.
>> 
>> Signed-off-by: H Hartley Sweeten <hsweeten at visionengravers.com>
>> Cc: David Brownell <david-b at pacbell.net>
>> Cc: linux-mtd at lists.infradead.org
>> 
>> ---
>> 
>> diff --git a/drivers/mtd/devices/mtd_dataflash.c b/drivers/mtd/devices/mtd_dataflash.c
>> index 93e3627..1981740 100644
>> --- a/drivers/mtd/devices/mtd_dataflash.c
>> +++ b/drivers/mtd/devices/mtd_dataflash.c
>> @@ -636,6 +636,7 @@ add_dataflash_otp(struct spi_device *spi, char *name,
>>  	struct mtd_info			*device;
>>  	struct flash_platform_data	*pdata = spi->dev.platform_data;
>>  	char				*otp_tag = "";
>> +	int				err = 0;
>>  
>>  	priv = kzalloc(sizeof *priv, GFP_KERNEL);
>>  	if (!priv)
>> @@ -693,13 +694,23 @@ add_dataflash_otp(struct spi_device *spi, char *name,
>>  
>>  		if (nr_parts > 0) {
>>  			priv->partitioned = 1;
>> -			return add_mtd_partitions(device, parts, nr_parts);
>> +			err = add_mtd_partitions(device, parts, nr_parts);
>> +			goto out;
>>  		}
>>  	} else if (pdata && pdata->nr_parts)
>>  		dev_warn(&spi->dev, "ignoring %d default partitions on %s\n",
>>  				pdata->nr_parts, device->name);
>>  
>> -	return add_mtd_device(device) == 1 ? -ENODEV : 0;
>> +	if (add_mtd_device(device) == 1)
>> +		err = -ENODEV;
>
> But if you fail here, you should also call del_mtd_partitions(), right?

Not as I understand it.

If the device has partitions (mtd_has_partitions), and the subsystem can
determine what they are, add_mtd_partitions is called to add those partitions.
The only way the code gets to add_mtd_device is if mtd_has_partitions returns
false or the number of partitions cannot be determined.  In that case the entire
device is added.  So calling del_mtd_partitions in that case is not valid.

Did I overlook something?

Regards,
Hartley