Re: [PATCH v5 13/31] qcow2: Update get/set_l2_entry() and add get/set_l2_bitmap()

[Alberto Garcia]

On Tue 05 May 2020 10:04:32 PM CEST, Eric Blake wrote:
> What happens for an image whose size is not cluster-aligned?  Must the
> bits corresponding to subclusters not present in the final cluster
> always be zero, or are they instead ignored regardless of value?

Attempting to read or write beyond the end of the image returns an error
(see blk_check_byte_request()).

But let's say we have a 32k image (only 1/2 of the first cluster is
used) and we manipulate the bitmap to mark all subclusters as allocated.

In this case if you try 'read 0 32k' then count_contiguous_subclusters()
would indeed say that there are 64k of data available. However the
caller knows that we only want 32k and

    if (bytes_available > bytes_needed) {
        bytes_available = bytes_needed;
    }

so in the end it doesn't really matter.

I think in practice it's the same as with traditional qcow2 images: once
we reach qcow2_get_host_offset() the code does not really know or care
about the size of the image or how much of the last cluster contains
actual data. It only cares about how much data the caller needs. The
limits have already been checked before.

But we could document it: "if the image size is not a multiple of the
cluster size then the bits corresponding to the subclusters beyond the
end of the image are ignored and should be set to zero", or something
like that.

Berto