On Mon, Sep 10, 2012 at 11:23 PM, Kevin Wolf <kw...@redhat.com> wrote:
> Am 10.08.2012 17:39, schrieb Dong Xu Wang:
>> Document for add-cow format, the usage and spec of add-cow are introduced.
>>
>> Signed-off-by: Dong Xu Wang <wdon...@linux.vnet.ibm.com>
>> ---
>> docs/specs/add-cow.txt | 123
>> ++++++++++++++++++++++++++++++++++++++++++++++++
>> 1 files changed, 123 insertions(+), 0 deletions(-)
>> create mode 100644 docs/specs/add-cow.txt
>>
>> diff --git a/docs/specs/add-cow.txt b/docs/specs/add-cow.txt
>> new file mode 100644
>> index 0000000..d5a7a68
>> --- /dev/null
>> +++ b/docs/specs/add-cow.txt
>> @@ -0,0 +1,123 @@
>> +== General ==
>> +
>> +The raw file format does not support backing files or copy on write feature.
>> +The add-cow image format makes it possible to use backing files with raw
>> +image by keeping a separate .add-cow metadata file. Once all sectors
>> +have been written into the raw image it is safe to discard the .add-cow
>> +and backing files, then we can use the raw image directly.
>> +
>> +An example usage of add-cow would look like::
>> +(ubuntu.img is a disk image which has been installed OS.)
>> + 1) Create a raw image with the same size of ubuntu.img
>> + qemu-img create -f raw test.raw 8G
>> + 2) Create an add-cow image which will store dirty bitmap
>> + qemu-img create -f add-cow test.add-cow \
>> + -o backing_file=ubuntu.img,image_file=test.raw
>> + 3) Run qemu with add-cow image
>> + qemu -drive if=virtio,file=test.add-cow
>> +
>> +test.raw may be larger than ubuntu.img, in that case, the size of
>> test.add-cow
>> +will be calculated from the size of test.raw.
>> +
>> +=Specification=
>> +
>> +The file format looks like this:
>> +
>> + +---------------+-------------+-----------------+
>> + | Header | Reserved | COW bitmap |
>> + +---------------+-------------+-----------------+
>> +
>> +All numbers in add-cow are stored in Little Endian byte order.
>> +
>> +== Header ==
>> +
>> +The Header is included in the first bytes:
>> +(#define HEADER_SIZE (4096 * header_pages_size))
>> + Byte 0 - 7: magic
>> + add-cow magic string ("ADD_COW\xff").
>> +
>> + 8 - 11: version
>> + Version number (only valid value is 1 now).
>> +
>> + 12 - 15: backing file name offset
>> + Offset in the add-cow file at which the backing file
>> + name is stored (NB: The string is not
>> nul-terminated).
>> + If backing file name does NOT exist, this field
>> will be
>> + 0. Must be between 80 and [HEADER_SIZE - 2](a file
>> name
>> + must be at least 1 byte).
>> +
>> + 16 - 19: backing file name size
>> + Length of the backing file name in bytes. It will
>> be 0
>> + if the backing file name offset is 0. If backing
>> file
>> + name offset is non-zero, then it must be non-zero.
>> Must
>> + be less than [HEADER_SIZE - 80] to fit in the
>> reserved
>> + part of the header.
>> +
>> + 20 - 23: image file name offset
>> + Offset in the add-cow file at which the image file
>> name
>> + is stored (NB: The string is not null terminated).
>> It
>> + must be between 80 and [HEADER_SIZE - 2].
>> +
>> + 24 - 27: image file name size
>> + Length of the image file name in bytes.
>> + Must be less than [HEADER_SIZE - 80] to fit in the
>> reserved
>> + part of the header.
>> +
>> + 28 - 35: features
>> + Currently only 1 feature bit is used:
>
> What happens when opening a file with an unknown bit set? How must
> unknown bits be initialised?
Okay, I will code as qcow2, report report_unsupported_feature error.
And I will update
the spec file.
>
>> + Feature bits:
>> + * ADD_COW_F_All_ALLOCATED = 0x01.
>
> What does this flag mean, and is it required to be set on that
> condition? Also, please use ALL_CAPS.
This feature bit will used as:
qemu-img create -f add-cow -o image_file=t.raw t.add-cow.
While creating add-cow and without backing_file, this feature can
avoid reading/updating
bitmap. I think it can let the code be more faster.
And also, maybe, I can implement add_cow_check, check if the feature
bit should be set.
How do you think, Kevin?
>
>> +
>> + 36 - 43: optional features
>> + Not used now. Reserved for future use. It must be
>> set to 0.
>
> And must be ignored when reading.
>
Okay.
>> +
>> + 44 - 47: header pages size
>> + The header field is variable-sized. This field
>> indicates
>> + how many pages(4k) will be used to store add-cow
>> header.
>> + In add-cow v1, it is fixed to 1, so the header size
>> will
>> + be 4k * 1 = 4096 bytes.
>
> Why arbitrarily defined "pages" instead of bytes or at least clusters?
Okay, next version I will just caclulate it by bytes.
>
>> +
>> + 48 - 63: backing file format
>> + format of backing file. It will be filled with 0 if
>> + backing file name offset is 0. If backing file name
>> + offset is non-zero, it must be non-zero. It is coded
>> + in free-form ASCII, and is not NUL-terminated.
>
> Zero padded on the right, I guess?
Yes, will update.
>
> Also defining that a string must be "non-zero" looks odd, should
> probably be "non-empty".
>
Okay.
>> +
>> + 64 - 79: image file format
>> + format of image file. It must be non-zero. It is
>> coded
>> + in free-form ASCII, and is not NUL-terminated.
>
> Same here.
Okay.
>
>> +
>> + 80 - [HEADER_SIZE - 1]:
>> + It is used to make sure COW bitmap field starts at
>> the
>> + HEADER_SIZE byte, backing file name and image file
>> name
>> + will be stored here. The bytes that is not pointing
>> to
>> + backing file and image file names will bet set to 0.
>
> "will be set to 0" describes the behaviour of qemu. A spec should
> describe the file format, not a specific implementation. Make it "must"
> or "should".
Okay.
>
>> +
>> +== COW bitmap ==
>> +
>> +The "COW bitmap" field starts at offset HEADER_SIZE, stores a bitmap
>> related to
>> +backing file and image file. The bitmap will track whether the sector in
>> +backing file is dirty or not.
>> +
>> +Each bit in the bitmap indicates one cluster's status. One cluster includes
>> 128
>> +sectors, then each bit indicates 512 * 128 = 64k bytes.
>
> Should we make the cluster size configurable?
>
>> the size of bitmap is
>> +calculated according to virtual size of image file, and it also should be
>> multipe
>
> Typo: multiple
>
> Sure you mean "should", or should it be "must"?
Okay.
>
>> +of 65536, the bits not used will be set to 0. Within each byte, the least
>> +significant bit covers the first cluster. Bit orders in one byte look like:
>> + +----+----+----+----+----+----+----+----+
>> + | b7 | b6 | b5 | b4 | b3 | b2 | b1 | b0 |
>> + +----+----+----+----+----+----+----+----+
>> +
>> +If the bit is 0, indicates the sector has not been allocated in image file,
>> data
>> +should be loaded from backing file while reading; if the bit is 1,
>> indicates the
>> +related sector has been dirty, should be loaded from image file while
>> reading.
>> +Writing to a sector causes the corresponding bit to be set to 1.
>> +
>> +If raw image is not an even multiple of cluster bytes, bits that correspond
>> to
>> +bytes beyond the raw file size in add-cow will be 0.
>
> "must be written as 0 and must be ignored when reading" or something
> like that.
Okay.
>
>> +Image file name and backing file name must NOT be the same, we prevent this
>> +while creating add-cow files.
>
> What we do is irrelevant for a spec.
Okay.
>
>> +Image file and backing file are interpreted relative to the qcow2 file, not
>> +to the current working directory of the process that opened the qcow2 file.
>
> Kevin
>
Thank you, Kevin.
