Re: file_chmod in unionfs, copy-on-write

[Thomas Bushnell, BSG]

Moritz Schulte <moritz@duesseldorf.ccc.de> writes:

>   just out of curiosity: how should the yet-to-be-written unionfs
> implement copy-on-write for file_chmod()?
> 
>   unionfs usually gives out ports to the underlying files; when in
> copy-on-write mode, it should give out a port to which the user can
> safely write without modifying the `real' data.  In my understanding
> this works, since unionfs notices at open time that the user intents
> to write to the file (the O_WRITE flag is specified).
> 
>   But for file_chmod there is no O_WRITE needed, therefore unionfs is
> not able to detect this `write'.

Well, copy-on-write is a nice idea, but it's not an inherent part of
unionfs.

If you want to catch file_* calls, then you really have to intervene
in every call; you can't hand out the real ports at all.