mendersoftware / mender / 1054468625

58%

master: 80%

Pull #1452

gitlab-ci

fix: Fix possible missing `Sync` call at the very end of the update.

We need to override the `Close` method so that we call `Sync` as the
final thing before closing. Most boards are not affected by this,
because the flushing interval is set to the native sector size of the
board, which, AFAIK, is 512 for all normal block devices except raw
Flash storage. I consider it very unlikely that a filesystem would not
be a multiple of 512 (maybe squashfs?), which means that by the time
we get to `Close()`, `Sync()` will usually have been called already
because the last write was 512 bytes or bigger.

But on raw Flash filesystems it can potentially happen, if the native
sector size is big, and the filesystem is not a multiple of this
number.

Changelog: Fix a rare bug which could corrupt the very end of a
rootfs-image update on a sudden powerloss. The circumstances where it
could happen are quite specific: The filesystem size in the update
need to *not* be a multiple of the native sector size, which is very
uncommon. The sector size is typically 512 bytes almost everywhere,
and hence filesystem also follow this block size, if not bigger. The
exception is raw Flash/UBI devices, where the sector size can be much
larger, and not a power of two, and hence these platforms may be more
susceptible.

Ticket: None

Signed-off-by: Kristian Amlie <kristian.amlie@northern.tech>
(cherry picked from commit ff0db6042)

7 of 8 new or added lines in 1 file covered. (87.5%)

6359 of 10930 relevant lines covered (58.18%)

30.99 hits per line