On Tue, Mar 14, 2017 at 9:46 AM, Filip Štědronský <r.lkml@xxxxxxxxxx> wrote: > Hi, > > On Tue, Mar 14, 2017 at 12:40:56PM +0200, Amir Goldstein wrote: >> An I am very happy that you used filehandles to keep track of objects >> in your example, because it fits my proposal really well. > > you can read more about the rationales in the WIP draft of my thesis > (especially the chapter Identifying Inodes and its surroundings): > > http://regnarg.cz/tmp/thesis.pdf > https://github.com/regnarg/bc > Filip, I read your draft and it's very interesting. Looking forward for the unfinished chapters ;-) open_by_handle() is important for your implementation in order to be able to get inode number (unique object id) from the file handle. In practice, although file handles are opaque descriptors, they usually contain the inode number information is a well known format. If it will be possible to take those formats and make them standard ABI, or alternatively, to call a new get_inum_by_handle() syscall, then the indexer/file sync won't need to have the extra privileges needed for open_by_handle() and file systems that can provide "view only" file handles (e.g. overlayfs) could be used with the indexer. Apropos chapter 1.3 "Filesystem based change detection" It's on my todo list to try and implement something with XFS using the LSN information embedded in every inode. It would be great if we can collaborate on this in the future. Amir.
