Store ownCloud's fileid in extended attributes

[PVince81]

If the underlying file system, local or external, supports extended attributes, how about storing the ownCloud fileid there ?

This would make it possible to:

• discover remote renames/moves: externally renaming a file/folder in a files_external storage breaks the contained shares  #11533
• restore oc_filecache entries when rescanning files, in case of DB crash https://github.com/owncloud/enterprise/issues/1167 and still be able to remap them to the correct metadata

@DeepDiver1975 @MorrisJobke @icewind1991 @Xenopathic

[PVince81]

also @karlitschek @rullzer @butonic

[PVince81]

I remember chatting with @schiesbn about this and he pointed out, what happens if two files report the same "fileid" from extended attributes. Not sure if that could ever happen, unless someone duplicated the file locally and the duplicate contains the same extended attrs.

[karlitschek]

yes. would be good if supported by the storage

[RobinMcCorkell]

We can in fact generalize the entire thing, and store all information that is currently stored in the filecache in extended attributes. Then we offload the heavy lifting from the DB layer to the filesystem layer, which may be more performant in certain circumstances.

[butonic]

I feel that we come across fs specific solutions for different problems. As @karlitschek already mentioned this won't work on all fs. Nevertheless, It does solve some problems. So I am for it.

[PVince81]

The Arch Wiki says that extended attributes are not copied by default with cp or rsync: https://wiki.archlinux.org/index.php/File_permissions_and_attributes#User_extended_attributes

So a situation where a duplicate fileid is found is less likely to happen. Still, need to decide what to do when found twice.

[PVince81]

Ah! It seems that @icewind1991 already had a POC for this: #14546

[ghost]

adding to the 9.1 discussion pile

[PVince81]

@jvillafanez some food for thought, in case you haven't thought of ext attributes before

[PVince81]

moving to backlog for now