libfdisk chooses a grain of 1MB fairly arbitrarily, and this granule
may not be honored by other utilities. GPT disks formatted elsewhere
may have space before the first partition, AND a partition that exists
solely below 1MB. If this occurs, cfdisk ends up adding a free space
region where end < start, resulting in a 16 Exabyte free region.
That's too many exabytes.
This happens because the start gets rounded up to the granule size in
new_freespace() but the end is left alone. The logs show it best:
23274: libfdisk: CXT: [0x572d878]: initialized: last=34, grain=2048
23274: libfdisk: CXT: [0x572d878]: partno=10, start=64
23274: libfdisk: CXT: [0x572d878]: freespace analyze: partno=10, start=64, end=64
23274: libfdisk: CXT: [0x572d878]: LBA 34 aligned-up 2048 [grain=2048s]
23274: libfdisk: CXT: [0x572d878]: LBA 63 aligned-down 0 [grain=2048s]
23274: libfdisk: CXT: [0x572d878]: LBA 34 aligned-near 0 [grain=2048s]
23274: libfdisk: CXT: [0x572d878]: 0 in range <2048..0> aligned to 2048
23274: libfdisk: PART: [0x574bb98]: alloc
23274: libfdisk: TAB: [0x5749d58]: adding freespace
23274: libfdisk: TAB: [0x5749d58]: insert entry 0x574bb98 pre=0x574a820 [start=2048, end=63, size=18446744073709549632, freespace ]
Avoid this by aligning the last value like new_freespace() does.
Signed-off-by: Evan Green <evangreen86@gmail.com>