Fix: update Hashfile function to return mtime and size as well.

[varmint708]

Possible race condition when the file is being updated by different process while simultaneously being checksummed by chkbit. (discovered and tested on macos, gave quite a scare there, might need some testing on other platforms)

[laktak]

Thanks for looking into this but AFAIK this is not a problem you can actually fix, at least not in this way.

Backups expect you not to write to the files you are backing up. If you do you will get inconsistent data. Imagine you backup part 1 of a database, then someone writes to part 1 before you backup the rest. This is not much different for chkbit when we calculate the checksum.

For backups on modern filesystems this can be solved with snapshots. I haven't looked into doing this with chkbit.

How did you test your fix?

One workaround would maybe getting the mtime before and after calculating the checksum and then rerun the checksum if the mtime has changed. Though even then it depends a lot on the filesystem and how often the mtime is updated.

[varmint708]

Thanks for looking into this but AFAIK this is not a problem you can actually fix, at least not in this way.

you are right, looking at it again, it doesn't fix it. anyways will cancel this PR.

Backups expect you not to write to the files you are backing up. If you do you will get inconsistent data. Imagine you backup part 1 of a database, then someone writes to part 1 before you backup the rest. This is not much different for chkbit when we calculate the checksum.

Agreed, i have a few local processes writing simultaneously, i should probably re-think my design there.

For backups on modern filesystems this can be solved with snapshots. I haven't looked into doing this with chkbit.

ok quickly tested this theory, used tmutil snapshot and mounted it in /tmp/ and did chkbit, that solved it for now. i checked backup util and they also prefer snapshot first, so could be a good idea to add this :). Thanks.

How did you test your fix?

i have local process which continuously writes to disk in a specific folder. if i run chkbit on this folder it returns DMG sometimes, but if i rsync it to another folder first and run chkbit there, all is good, or even running on backup disk shows no damage, so test was simple, rerun chkbit in this folder 10~12 times to see if it shows DMG.

One workaround would maybe getting the mtime before and after calculating the checksum and then rerun the checksum if the mtime has changed. Though even then it depends a lot on the filesystem and how often the mtime is updated.

Agreed, this may alleviate the problem a bit, so we can try this before snapshotting.