Commit 9b56d543

 ("dump_skip(): dump_seek() replacement taking
coredump_params") introduced a regression with regard to RLIMIT_CORE.
Previously, when a core dump was sparse, only the data that was actually
written out would count against the limit. Now, the sparse ranges are
also included, which leads to truncated core dumps when the actual disk
usage is still well below the limit. Restore the old behavior by only
counting what gets emitted and ignoring what gets skipped.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>