Re: [PATCH 1/2] vfs: replace i_readcount with a biased i_count

From: Amir Goldstein <amir73il@gmail.com>
To: Mimi Zohar <zohar@linux.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>,
	"J . Bruce Fields" <bfields@fieldses.org>,
	Jeff Layton <jlayton@poochiereds.net>,
	Al Viro <viro@zeniv.linux.org.uk>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>,
	overlayfs <linux-unionfs@vger.kernel.org>,
	linux-integrity <linux-integrity@vger.kernel.org>
Subject: Re: [PATCH 1/2] vfs: replace i_readcount with a biased i_count
Date: Wed, 12 Jun 2019 18:09:59 +0300	[thread overview]
Message-ID: <CAOQ4uxhooVwtHcDCr4hu+ovzKGUdWfQ+3F3nbgK3HXgV+fUK9w@mail.gmail.com> (raw)
In-Reply-To: <1560343899.4578.9.camel@linux.ibm.com>

On Wed, Jun 12, 2019 at 3:52 PM Mimi Zohar <zohar@linux.ibm.com> wrote:
>
> On Sat, 2019-06-08 at 16:57 +0300, Amir Goldstein wrote:
> > Count struct files open RO together with inode reference count instead
> > of using a dedicated i_readcount field.  This will allow us to use the
> > RO count also when CONFIG_IMA is not defined and will reduce the size of
> > struct inode for 32bit archs when CONFIG_IMA is defined.
> >
> > We need this RO count for posix leases code, which currently naively
> > checks i_count and d_count in an inaccurate manner.
> >
> > Should regular i_count overflow into RO count bias by struct files
> > opened for write, it's not a big deal, as we mostly need the RO count
> > to be reliable when the first writer comes along.
>
> "i_count" has been defined forever.  Has its meaning changed?  This
> patch implies that "i_readcount" was never really needed.
>

Not really.
i_count is only used to know if object is referenced.
It does not matter if user takes 1 or more references on i_count
as long as user puts back all the references it took.

If user took i_readcount, i_count cannot be zero, so short of overflow,
we can describe i_readcount as a biased i_count.

But if I am following Miklos' suggestion to make i_count 64bit, inode
struct size is going to grow for 32bit arch when  CONFIG_IMA is not
defined, so to reduce impact, I will keep i_readcount as a separate
member and let it be defined also when BITS_PER_LONG == 64
and implement inode_is_open_rdonly() using d_count and i_count
when i_readcount is not defined.

Let's see what people will have to say about that...

Thanks,
Amir.