[Qemu-devel] [PATCH 03/12] VMDK: probe for mono flat image

[Qemu-devel] [PATCH 03/12] VMDK: probe for mono flat image

[Fam Zheng]

vmdk_probe for mono flat images.

Signed-off-by: Fam Zheng <famcool@gmail.com>
---
 block/vmdk.c |   13 ++++++++++---
 1 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/block/vmdk.c b/block/vmdk.c
index f787528..bf8d02a 100644
--- a/block/vmdk.c
+++ b/block/vmdk.c
@@ -101,10 +101,17 @@ static int vmdk_probe(const uint8_t *buf, int
buf_size, const char *filename)
         return 0;
     magic = be32_to_cpu(*(uint32_t *)buf);
     if (magic == VMDK3_MAGIC ||
-        magic == VMDK4_MAGIC)
+        magic == VMDK4_MAGIC) {
         return 100;
-    else
-        return 0;
+    } else {
+        char *cid_p, *ct_p, *extent_p;
+        cid_p = strstr((char *)buf, "CID");
+        ct_p = strstr((char *)buf, "createType");
+        extent_p = strstr((char *)buf, "RW");
+        if (cid_p && ct_p && extent_p)
+            return 100;
+    }
+    return 0;
 }

 #define CHECK_CID 1

Re: [Qemu-devel] [PATCH 03/12] VMDK: probe for mono flat image

[Stefan Hajnoczi]

On Sat, Jun 04, 2011 at 08:40:50AM +0800, Fam Zheng wrote:
> vmdk_probe for mono flat images.
> 
> Signed-off-by: Fam Zheng <famcool@gmail.com>
> ---
>  block/vmdk.c |   13 ++++++++++---
>  1 files changed, 10 insertions(+), 3 deletions(-)
> 
> diff --git a/block/vmdk.c b/block/vmdk.c
> index f787528..bf8d02a 100644
> --- a/block/vmdk.c
> +++ b/block/vmdk.c
> @@ -101,10 +101,17 @@ static int vmdk_probe(const uint8_t *buf, int
> buf_size, const char *filename)
>          return 0;
>      magic = be32_to_cpu(*(uint32_t *)buf);
>      if (magic == VMDK3_MAGIC ||
> -        magic == VMDK4_MAGIC)
> +        magic == VMDK4_MAGIC) {
>          return 100;
> -    else
> -        return 0;
> +    } else {
> +        char *cid_p, *ct_p, *extent_p;
> +        cid_p = strstr((char *)buf, "CID");
> +        ct_p = strstr((char *)buf, "createType");
> +        extent_p = strstr((char *)buf, "RW");
> +        if (cid_p && ct_p && extent_p)
> +            return 100;

NUL-terminated string functions cannot be used for probing because the
input file may be invalid.  If the magic number matches but there is no
NUL in the buffer then the strstr(3) will run off the end of the buffer.

Also note that the specification says "The descriptor file is not
case-sensitive".  "cid", "CiD", and "CID" should all be allowed.

Do non-monolithic vmdk images always have "# Disk DescriptorFile" as the
first line?  Perhaps you can test for that using memcmp(3) instead.

Stefan

Re: [Qemu-devel] [PATCH 03/12] VMDK: probe for mono flat image

[Fam Zheng]

On Tue, Jun 14, 2011 at 4:06 PM, Stefan Hajnoczi <stefanha@gmail.com> wrote:
> On Sat, Jun 04, 2011 at 08:40:50AM +0800, Fam Zheng wrote:
>> vmdk_probe for mono flat images.
>>
>> Signed-off-by: Fam Zheng <famcool@gmail.com>
>> ---
>>  block/vmdk.c |   13 ++++++++++---
>>  1 files changed, 10 insertions(+), 3 deletions(-)
>>
>> diff --git a/block/vmdk.c b/block/vmdk.c
>> index f787528..bf8d02a 100644
>> --- a/block/vmdk.c
>> +++ b/block/vmdk.c
>> @@ -101,10 +101,17 @@ static int vmdk_probe(const uint8_t *buf, int
>> buf_size, const char *filename)
>>          return 0;
>>      magic = be32_to_cpu(*(uint32_t *)buf);
>>      if (magic == VMDK3_MAGIC ||
>> -        magic == VMDK4_MAGIC)
>> +        magic == VMDK4_MAGIC) {
>>          return 100;
>> -    else
>> -        return 0;
>> +    } else {
>> +        char *cid_p, *ct_p, *extent_p;
>> +        cid_p = strstr((char *)buf, "CID");
>> +        ct_p = strstr((char *)buf, "createType");
>> +        extent_p = strstr((char *)buf, "RW");
>> +        if (cid_p && ct_p && extent_p)
>> +            return 100;
>
> NUL-terminated string functions cannot be used for probing because the
> input file may be invalid.  If the magic number matches but there is no
> NUL in the buffer then the strstr(3) will run off the end of the buffer.
>
> Also note that the specification says "The descriptor file is not
> case-sensitive".  "cid", "CiD", and "CID" should all be allowed.
>
> Do non-monolithic vmdk images always have "# Disk DescriptorFile" as the
> first line?  Perhaps you can test for that using memcmp(3) instead.
No guarantee in specification, although VMware does start descriptors
with such a line "# Disk DescriptorFile".
But is it proper that we make this assumption?


-- 
Best regards!
Fam Zheng

Re: [Qemu-devel] [PATCH 03/12] VMDK: probe for mono flat image

[Stefan Hajnoczi]

On Wed, Jun 15, 2011 at 5:26 AM, Fam Zheng <famcool@gmail.com> wrote:
> On Tue, Jun 14, 2011 at 4:06 PM, Stefan Hajnoczi <stefanha@gmail.com> wrote:
>> On Sat, Jun 04, 2011 at 08:40:50AM +0800, Fam Zheng wrote:
>>> vmdk_probe for mono flat images.
>>>
>>> Signed-off-by: Fam Zheng <famcool@gmail.com>
>>> ---
>>>  block/vmdk.c |   13 ++++++++++---
>>>  1 files changed, 10 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/block/vmdk.c b/block/vmdk.c
>>> index f787528..bf8d02a 100644
>>> --- a/block/vmdk.c
>>> +++ b/block/vmdk.c
>>> @@ -101,10 +101,17 @@ static int vmdk_probe(const uint8_t *buf, int
>>> buf_size, const char *filename)
>>>          return 0;
>>>      magic = be32_to_cpu(*(uint32_t *)buf);
>>>      if (magic == VMDK3_MAGIC ||
>>> -        magic == VMDK4_MAGIC)
>>> +        magic == VMDK4_MAGIC) {
>>>          return 100;
>>> -    else
>>> -        return 0;
>>> +    } else {
>>> +        char *cid_p, *ct_p, *extent_p;
>>> +        cid_p = strstr((char *)buf, "CID");
>>> +        ct_p = strstr((char *)buf, "createType");
>>> +        extent_p = strstr((char *)buf, "RW");
>>> +        if (cid_p && ct_p && extent_p)
>>> +            return 100;
>>
>> NUL-terminated string functions cannot be used for probing because the
>> input file may be invalid.  If the magic number matches but there is no
>> NUL in the buffer then the strstr(3) will run off the end of the buffer.
>>
>> Also note that the specification says "The descriptor file is not
>> case-sensitive".  "cid", "CiD", and "CID" should all be allowed.
>>
>> Do non-monolithic vmdk images always have "# Disk DescriptorFile" as the
>> first line?  Perhaps you can test for that using memcmp(3) instead.
> No guarantee in specification, although VMware does start descriptors
> with such a line "# Disk DescriptorFile".
> But is it proper that we make this assumption?

Have you tried removing that comment line to see if VMware still
recognizes the file?

Stefan

Re: [Qemu-devel] [PATCH 03/12] VMDK: probe for mono flat image

[Fam Zheng]

> Have you tried removing that comment line to see if VMware still
> recognizes the file?
>

Yes, it recognizes iff the first option line (non-comment, non-empty)
is "version=1" or "version=2“.

(An interesting thing that it is both space sensitive and case sensitive)


-- 
Best regards!
Fam Zheng

Re: [Qemu-devel] [PATCH 03/12] VMDK: probe for mono flat image

[Stefan Hajnoczi]

On Wed, Jun 15, 2011 at 6:45 AM, Fam Zheng <famcool@gmail.com> wrote:
>> Have you tried removing that comment line to see if VMware still
>> recognizes the file?
>>
>
> Yes, it recognizes iff the first option line (non-comment, non-empty)
> is "version=1" or "version=2“.
>
> (An interesting thing that it is both space sensitive and case sensitive)

Cool, nice job figuring this out.  We should do the same.

Stefan