Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 9 May 2023 14:53:41 -0400
From:   "Theodore Ts'o" <tytso@mit.edu>
To:     Matthew Wilcox <willy@infradead.org>
Cc:     Tudor Ambarus <tudor.ambarus@linaro.org>, adilger.kernel@dilger.ca,
        linux-ext4@vger.kernel.org, linux-kernel@vger.kernel.org,
        joneslee@google.com, linux-mm@kvack.org
Subject: Re: [PATCH] ext4: remove superfluous check that pointer is not NULL
Message-ID: <ZFqWtcmRwxhyem4p@mit.edu>
References: <20230508151337.79304-1-tudor.ambarus@linaro.org>
 <ZFkf/oJnCLZSWgYr@mit.edu>
 <ZFll93wsEUZIV/aI@casper.infradead.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZFll93wsEUZIV/aI@casper.infradead.org>
Precedence: bulk

On Mon, May 08, 2023 at 10:13:27PM +0100, Matthew Wilcox wrote:
> > 
> > I was looking at this just a few weeks ago, and I couldn't find any
> > actual *documentation* that it was safe to call vfree(NIILL) or
> > kvfree(NULL).  The problem is there are a lot of architecture-specific
> > functions, and unlike with kfree() there is no top-level "if (ptr ==
> > NULL) return;" in the top-level vfree() and kvfree().
> 
> There doesn't need to be in kvfree().  is_vmalloc_addr() returns 'false'
> for NULL, so it calls kfree(), which as you note has an explicit check
> for ZERO_OR_NULL_PTR().  is_vmalloc_addr() also returns false for the
> ZERO pointer, fwiw.
> 
> I agree that this should be explicitly documented as allowed, since it's
> not reasonable to expect users to dig through these functions to verify
> that such a change is safe.

I seem to recall at one point looking at kvfree_rcu (at least the one
argument variant), and I *thought* it would unconditionally allocate
memory so it could be put on a linked list to be freed after an RCU
grace period had elapsed.  But I tried tracing through the huge
numbers of cpp macros and other layers of #ifdef's and other
abstractions, and in my conference-induced sleep depreviation, it
caused my head to spin, and I gave up trying to trace it down so I had
100% confidence.

So if someone could document *all* of the k[v]free_* variants whether
it is safe/optimal to pass NULL to them, that would be great, thanks.

						- Ted