cow_brd read unmodified data from parent in snapshot disk fix

[ashmrtn]

Stems from a copy/paste error. If a range of data spanning multiple pages was requested from a disk snapshot and the snapshot had not modified any of that data (i.e., it was reading the data from the parent device) then an incorrect offset into the second page of data was used when copying data from the second page.

[vijay03]

Does this affect any of our results so far?

[ashmrtn]

We were discussing this and we don't think so. The chances of hitting it are actually fairly low since it requires:

• file system with < 4k block size
• disk snapshot to not have modified the blocks in question
• read that is not 4k aligned

if anything, it would seem like if we were hitting this bug we would be getting confusing results. I also believe that the bulk of our tests up until this point have either not modified file data directly (i.e., unaligned fallocate which often works with the extent tree which would likely be 4k-aligned and/or already modified by some other operation) or have been 4k-aligned operations.

[vijay03]

Sounds good!