Re: [PATCH] [DRAFT RFC]: file: reclaim 24 bytes from f_owner

[Al Viro <viro@zeniv.linux.org.uk>]

On Fri, Aug 09, 2024 at 10:10:40PM +0200, Christian Brauner wrote:

> fcntl()s and file leases can just allocate on demand easily. Cleanup
> happens during __fput() when file was really opend. For fcntl()s and
> file leases this is guaranteed because the file is already alive. For
> drivers they need to cleanup the allocated memory before they've
> succesfully finished ->open(). Afterwards we'll just clean it up.
> 
> Interactions with O_PATH should be fine as well e.g., when opening a
> /dev/tty as O_PATH then no ->open() happens thus no filp->f_owner is
> allocated. That's fine as no file operation will be set for those and
> the device has never been opened. fcntl()s called on such things will
> just allocate a ->f_owner on demand. Although I have zero idea why'd you
> care about f_owner on an O_PATH fd.

One general note: IMO you are far too optimistic about the use of __cleanup
extensions; it's _not_ something that we want blindly used all over
the place.  In some cases it's fine, but I'm very nervous about the
possibility of people starting to cargo-cult it all over the place.

Again, __cleanup support in gcc has significant holes, at least up to
gcc 12.

This is *NOT* a generally safe part of C dialect we are using.  And
the pitfalls associated with it are not documented, let alone generally
understood.