block-cloning-tool

a script made for helping handle block cloning files with-out some of the footguns from using GNU coreutils cp

Known issues

Linux only right now. Also only will work with posix ACLs and not NFSv4 ACLs, this is an API issue that I need to investigate

Issues that cp has

• It will wipe out ACLs and xattr on the cloned file
• Potential race conditions where the file could be changed out before cp does the operation
• Resets [amcb]times on the files.

... others?

How this works

First it will do some sanity checks to be sure that it's not going to make a clone that shouldn't be done. This is to check for races between finding duplicate files with rmlint, fdupes, or other tools and the time of attempting the clone

1. Start by flock()ing the source and destination files. Not perfect but it will help prevent something else from replacing the file out from under us.
2. Check that the lengths match on both source and destination files, mostly as a very quick sanity check before we do anything to the files
3. Check the digest of the files (BLAKE3?) to check that they're identical. (should it instead do a complete byte by byte comparison?)
4. Serialize the xattr, acl, and [amcb]times of the destination file
5. Create a new file (dest filename + mktmpfile-ish), setacls, xattr on the new file and flock it
6. make the ioctl(FIDEDUP) call to clone the file 6a) If that errors, and --across-datasets is provided try doing copyfilerange() syscall 6b) TODO any reasonable way to warn if we think that didn't clone?
7. set the times on the file (TODO how?)
8. rename old dest file
9. rename new cloned file
10. unlink ole dest file