Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 27 Aug 2020 15:14:37 -0700
From:   Kees Cook <keescook@chromium.org>
To:     Al Viro <viro@zeniv.linux.org.uk>
Cc:     Linus Torvalds <torvalds@linux-foundation.org>,
        James Bottomley <James.Bottomley@hansenpartnership.com>,
        Allen Pais <allen.lkml@gmail.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Jens Axboe <axboe@kernel.dk>
Subject: Re: [PATCH] linux/kernel.h: add container_from()
Message-ID: <202008271503.181A6A609@keescook>
References: <20200827013636.149307-1-allen.lkml@gmail.com>
 <CAHk-=whiEUUrtnbgUH2xsD0+jNyoXudYJ4hGCA55MCjryaHGjw@mail.gmail.com>
 <1598553133.4237.8.camel@HansenPartnership.com>
 <CAHk-=wi8o+FvfQkUiH_2MUs3J19FzfMzumOViAJ2aboGg9qY7Q@mail.gmail.com>
 <CAHk-=wingJWToQfoc+m2am7Q=7r8XD+6p0FXasCRAzOdcRyngw@mail.gmail.com>
 <202008271150.7231B901@keescook>
 <CAHk-=whDhHWQo_QjZp36=x=GLMGOJ2xnfsUk9xkUuWRz=i9gOg@mail.gmail.com>
 <20200827213636.GF1236603@ZenIV.linux.org.uk>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200827213636.GF1236603@ZenIV.linux.org.uk>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Aug 27, 2020 at 10:36:36PM +0100, Al Viro wrote:
> On Thu, Aug 27, 2020 at 01:46:33PM -0700, Linus Torvalds wrote:
> > You really have to pick some pretty excessive type names (or variable
> > names) to get close to 80 characters. Again, to pick an example:
> > 
> >         struct timer_group_priv *priv = container_of(handle,
> >                         struct timer_group_priv, timer[handle->num]);
> > 
> > ends up being long even if you were to split it, but that funky
> > container_from() wouldn't have helped the real problem - the fact that
> > the above is complex and nasty.

The point about doing the assignment with the declaration certainly makes
the "ugliness" worse, I agree. I'm still not generally convinced about
the redundancy level pros/cons, but I concede that having a common idiom
(rather than a succinct but subsystem-dependent idiom) is better for
people reading the code for the first time.

> > And I had to _search_ for that example. All the normal cases of
> > split-line container-of's were due to doing it with the declaration,
> > or beause the first argument ended up being an expression in itself
> > and the nested expressions made it more complex.
> 
> Speaking of searching, this kind of typeof use is, IMO, actively
> harmful - it makes finding the places where we might get from
> e.g. linked list to containing objects much harder.  container_of
> (unless combined with obfuscating use of typeof()) at least gives
> you a chance to grep - struct foo *not* followed by '*' is a pattern
> that doesn't give too many false positives.  This one, OTOH, is
> essentially impossible to grep for.

And this observation about workflow does strike a chord with me. I do end
up with those kind of searches too. In trying to examine my preferences
here, I think my instincts are to avoid open-coded types (leading me to
want to use typeof()) but I think those instincts were actually developed
from dealing with _sizeof_ and all the way it goes terribly wrong. So,
okay, I'm convinced. container_of() it is.

Doing these conversions becomes a little less mechanical if assignment
needs to be split from declaration, but hey, we've got a 100 character
line "limit" now, so maybe it'll be less needed. :)

-- 
Kees Cook