diff mbox

[v2] vmdk: return ERROR when cluster sector is larger than vmdk limitation

Message ID 20180322062447.16537-1-yuchenlin@synology.com (mailing list archive)
State New, archived
Headers show

Commit Message

Zhijian Li (Fujitsu)" via March 22, 2018, 6:24 a.m. UTC
From: yuchenlin <yuchenlin@synology.com>

VMDK has a hard limitation of extent size, which is due to the size of grain
table entry is 32 bits. It means it can only point to a grain located at
offset = 2^32. To avoid writing the user data beyond limitation and record a useless offset
in grain table. We should return ERROR here.

Signed-off-by: yuchenlin <yuchenlin@synology.com>
---
v1->v2:
 - change commit message
 - check before allocating
 - should be >=
 - the unit is sector now

 thanks

 block/vmdk.c | 6 ++++++
 1 file changed, 6 insertions(+)

Comments

Fam Zheng March 22, 2018, 7:41 a.m. UTC | #1
On Thu, 03/22 14:24, yuchenlin@synology.com wrote:
> From: yuchenlin <yuchenlin@synology.com>
> 
> VMDK has a hard limitation of extent size, which is due to the size of grain
> table entry is 32 bits. It means it can only point to a grain located at
> offset = 2^32. To avoid writing the user data beyond limitation and record a useless offset
> in grain table. We should return ERROR here.
> 
> Signed-off-by: yuchenlin <yuchenlin@synology.com>
> ---
> v1->v2:
>  - change commit message
>  - check before allocating
>  - should be >=
>  - the unit is sector now
> 
>  thanks
> 
>  block/vmdk.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/block/vmdk.c b/block/vmdk.c
> index f94c49a9c0..a1c21dbbba 100644
> --- a/block/vmdk.c
> +++ b/block/vmdk.c
> @@ -47,6 +47,8 @@
>  #define VMDK4_FLAG_MARKER (1 << 17)
>  #define VMDK4_GD_AT_END 0xffffffffffffffffULL
>  
> +#define VMDK_EXTENT_MAX_SECTORS (4294967296)
> +
>  #define VMDK_GTE_ZEROED 0x1
>  
>  /* VMDK internal error codes */
> @@ -1250,6 +1252,10 @@ static int get_cluster_offset(BlockDriverState *bs,
>              return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
>          }
>  
> +        if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
> +            return VMDK_ERROR;
> +        }
> +
>          cluster_sector = extent->next_cluster_sector;
>          extent->next_cluster_sector += extent->cluster_sectors;
>  
> -- 
> 2.16.2
> 

Cc'ing Kevin and Max who merge block/ patches.

Reviewed-by: Fam Zheng <famz@redhat.com>
Eric Blake March 22, 2018, 1:07 p.m. UTC | #2
On 03/22/2018 02:41 AM, Fam Zheng wrote:
> On Thu, 03/22 14:24, yuchenlin@synology.com wrote:
>> From: yuchenlin <yuchenlin@synology.com>
>>
>> VMDK has a hard limitation of extent size, which is due to the size of grain
>> table entry is 32 bits. It means it can only point to a grain located at
>> offset = 2^32. To avoid writing the user data beyond limitation and record a useless offset
>> in grain table. We should return ERROR here.
>>
>> Signed-off-by: yuchenlin <yuchenlin@synology.com>
>> ---

>> +++ b/block/vmdk.c
>> @@ -47,6 +47,8 @@
>>   #define VMDK4_FLAG_MARKER (1 << 17)
>>   #define VMDK4_GD_AT_END 0xffffffffffffffffULL
>>   
>> +#define VMDK_EXTENT_MAX_SECTORS (4294967296)
>> +

Hmm, with sectors instead of bytes, you don't want to use Phillipe's '2 
* T_BYTE' macro after all; but it would still look better spelled as 
'(1ULL << 32)' or '(4ULL * 1024 * 1024 * 1024)'.  Most importantly, 
since the value does NOT fit in a 32-bit unsigned integer, I think you 
NEED to use the ULL suffix to make your intent clear.

>>   #define VMDK_GTE_ZEROED 0x1
>>   
>>   /* VMDK internal error codes */
>> @@ -1250,6 +1252,10 @@ static int get_cluster_offset(BlockDriverState *bs,
>>               return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
>>           }
>>   
>> +        if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
>> +            return VMDK_ERROR;
>> +        }
>> +
>>           cluster_sector = extent->next_cluster_sector;
>>           extent->next_cluster_sector += extent->cluster_sectors;
>>   
>> -- 
>> 2.16.2
>>
> 
> Cc'ing Kevin and Max who merge block/ patches.

Also, CC'ing qemu-block, as all block-related patches should go through 
there.

> 
> Reviewed-by: Fam Zheng <famz@redhat.com>
> 
> 
>
diff mbox

Patch

diff --git a/block/vmdk.c b/block/vmdk.c
index f94c49a9c0..a1c21dbbba 100644
--- a/block/vmdk.c
+++ b/block/vmdk.c
@@ -47,6 +47,8 @@ 
 #define VMDK4_FLAG_MARKER (1 << 17)
 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
 
+#define VMDK_EXTENT_MAX_SECTORS (4294967296)
+
 #define VMDK_GTE_ZEROED 0x1
 
 /* VMDK internal error codes */
@@ -1250,6 +1252,10 @@  static int get_cluster_offset(BlockDriverState *bs,
             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
         }
 
+        if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
+            return VMDK_ERROR;
+        }
+
         cluster_sector = extent->next_cluster_sector;
         extent->next_cluster_sector += extent->cluster_sectors;