Thanks for handling this.
The fix looks good, and I can't see any other references
to the realloc'd memory that need to be updated.
That code is only enabled for sparse files at present
so maybe the NEWS might be adjusted slightly like:
cp could read from freed memory and could even make corrupt copies.
This could happen with a very fragmented and sparse input file,
on GNU/Linux file systems supporting fiemap extent scanning.
[bug introduced in coreutils-8.11]
That bug is obviously serious and the fix merits a prompt release.
What would you or anyone else like to see committed before the release?
Ondrej's df patch is overdue, but I haven't reviewed it, and now wonder
if it'd be better to defer it until after 8.20. Without reviewing it,
however, I'm not well placed to estimate its risk/benefit ratio.